Merge pull request #198 from ShowLo/patch-2

Fix the bug that causes an infinite loop when the total number of fra…

README.md

@@ -29,6 +29,7 @@ We introduce `MuseTalk`, a **real-time high quality** lip-syncing model (30fps+
 - [04/02/2024] Release MuseTalk project and pretrained models.
 - [04/16/2024] Release Gradio [demo](https://huggingface.co/spaces/TMElyralab/MuseTalk) on HuggingFace Spaces (thanks to HF team for their community grant)
 - [04/17/2024] :mega: We release a pipeline that utilizes MuseTalk for real-time inference.
+- [04/30/2024] We release an initial version of training codes in `train_codes`.
 
 ## Model
 ![Model Structure](assets/figs/musetalk_arc.jpg)
@@ -165,7 +166,7 @@ Note that although we use a very similar architecture as Stable Diffusion, MuseT
 - [x] Huggingface Gradio [demo](https://huggingface.co/spaces/TMElyralab/MuseTalk).
 - [x] codes for real-time inference.
 - [ ] technical report.
-- [ ] training codes.
+- [x] training codes.
 - [ ] a better model (may take longer).
 
 

data_new.sh

@@ -0,0 +1,77 @@
+#!/bin/bash
+
+# Function to extract video and audio sections
+extract_sections() {
+  input_video=$1
+  base_name=$(basename "$input_video" .mp4)
+  output_dir=$2
+  split=$3
+  duration=$(ffmpeg -i "$input_video" 2>&1 | grep Duration | awk '{print $2}' | tr -d ,)
+  IFS=: read -r hours minutes seconds <<< "$duration"
+  total_seconds=$((10#${hours}*3600 + 10#${minutes}*60 + 10#${seconds%.*}))
+  chunk_size=180  # 3 minutes in seconds
+  index=0
+
+  mkdir -p "$output_dir"
+
+  while [ $((index * chunk_size)) -lt $total_seconds ]; do
+    start_time=$((index * chunk_size))
+    section_video="${output_dir}/${base_name}_part${index}.mp4"
+    section_audio="${output_dir}/${base_name}_part${index}.mp3"
+    
+    ffmpeg -i "$input_video" -ss "$start_time" -t "$chunk_size" -c copy "$section_video"
+    ffmpeg -i "$input_video" -ss "$start_time" -t "$chunk_size" -q:a 0 -map a "$section_audio"
+    
+    # Create and update the config.yaml file
+    echo "task_0:" > config.yaml
+    echo "  video_path: \"$section_video\"" >> config.yaml
+    echo "  audio_path: \"$section_audio\"" >> config.yaml
+
+    # Run the Python script with the current config.yaml
+    python -m scripts.data --inference_config config.yaml --folder_name "$base_name"
+    
+    index=$((index + 1))
+  done
+
+  # Clean up save folder
+  rm -rf $output_dir
+}
+
+# Main script
+if [ $# -lt 3 ]; then
+  echo "Usage: $0 <train/test> <output_directory> <input_videos...>"
+  exit 1
+fi
+
+split=$1
+output_dir=$2
+shift 2
+input_videos=("$@")
+
+# Initialize JSON array
+json_array="["
+
+for input_video in "${input_videos[@]}"; do
+  base_name=$(basename "$input_video" .mp4)
+  
+  # Extract sections and run the Python script for each section
+  extract_sections "$input_video" "$output_dir" "$split"
+
+  # Add entry to JSON array
+  json_array+="\"../data/images/$base_name\","
+done
+
+# Remove trailing comma and close JSON array
+json_array="${json_array%,}]"
+
+# Write JSON array to the correct file
+if [ "$split" == "train" ]; then
+  echo "$json_array" > train.json
+elif [ "$split" == "test" ]; then
+  echo "$json_array" > test.json
+else
+  echo "Invalid split: $split. Must be 'train' or 'test'."
+  exit 1
+fi
+
+echo "Processing complete."

scripts/data.py

@@ -0,0 +1,257 @@
+import cv2
+import os
+# import dlib
+import argparse
+import os
+from omegaconf import OmegaConf
+import numpy as np
+import cv2
+import torch
+import glob
+import pickle
+from tqdm import tqdm
+import copy
+import uuid
+
+from musetalk.utils.utils import get_file_type,get_video_fps
+from musetalk.utils.preprocessing import get_landmark_and_bbox,read_imgs,coord_placeholder
+from musetalk.utils.blending import get_image
+from musetalk.utils.utils import load_all_model
+import shutil
+import gc
+
+# load model weights
+audio_processor, vae, unet, pe = load_all_model()
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+timesteps = torch.tensor([0], device=device)
+
+def get_largest_integer_filename(folder_path):
+    # Check if the folder exists
+    if not os.path.isdir(folder_path):
+        return -1
+
+    # Get the list of files in the folder
+    files = os.listdir(folder_path)
+
+    # Check if the folder is empty
+    if not files:
+        return -1
+
+    # Extract the integer part of filenames and find the largest
+    largest_integer = -1
+    for file in files:
+        try:
+            # Get the integer part of the filename
+            file_int = int(os.path.splitext(file)[0])
+            if file_int > largest_integer:
+                largest_integer = file_int
+        except ValueError:
+            # Skip files that don't have an integer filename
+            continue
+
+    return largest_integer
+
+def datagen(whisper_chunks,
+            crop_images,
+            batch_size=8,
+            delay_frame=0):
+    whisper_batch, crop_batch = [], []
+    for i, w in enumerate(whisper_chunks):
+        idx = (i+delay_frame)%len(crop_images)
+        crop_image = crop_images[idx]
+        whisper_batch.append(w)
+        crop_batch.append(crop_image)
+
+        if len(crop_batch) >= batch_size:
+            whisper_batch = np.stack(whisper_batch)
+            # latent_batch = torch.cat(latent_batch, dim=0)
+            yield whisper_batch, crop_batch
+            whisper_batch, crop_batch = [], []
+
+    # the last batch may smaller than batch size
+    if len(crop_batch) > 0:
+        whisper_batch = np.stack(whisper_batch)
+        # latent_batch = torch.cat(latent_batch, dim=0)
+
+        yield whisper_batch, crop_batch
+
+@torch.no_grad()
+def main(args):
+    global pe
+    if args.use_float16 is True:
+        pe = pe.half()
+        vae.vae = vae.vae.half()
+        unet.model = unet.model.half()
+    
+    inference_config = OmegaConf.load(args.inference_config)
+    total_audio_index=get_largest_integer_filename(f"data/audios/{args.folder_name}")
+    total_image_index=get_largest_integer_filename(f"data/images/{args.folder_name}")
+    temp_audio_index=total_audio_index
+    temp_image_index=total_image_index
+    for task_id in inference_config:
+        video_path = inference_config[task_id]["video_path"]
+        audio_path = inference_config[task_id]["audio_path"]
+        bbox_shift = inference_config[task_id].get("bbox_shift", args.bbox_shift)
+        folder_name = args.folder_name
+        if not os.path.exists(f"data/images/{folder_name}/"):
+            os.makedirs(f"data/images/{folder_name}")
+        if not os.path.exists(f"data/audios/{folder_name}/"):
+            os.makedirs(f"data/audios/{folder_name}")
+        input_basename = os.path.basename(video_path).split('.')[0]
+        audio_basename  = os.path.basename(audio_path).split('.')[0]
+        output_basename = f"{input_basename}_{audio_basename}"
+        result_img_save_path = os.path.join(args.result_dir, output_basename) # related to video & audio inputs
+        crop_coord_save_path = os.path.join(result_img_save_path, input_basename+".pkl") # only related to video input
+        os.makedirs(result_img_save_path,exist_ok =True)
+        
+        if args.output_vid_name is None:
+            output_vid_name = os.path.join(args.result_dir, output_basename+".mp4")
+        else:
+            output_vid_name = os.path.join(args.result_dir, args.output_vid_name)
+        ############################################## extract frames from source video ##############################################
+        if get_file_type(video_path)=="video":
+            save_dir_full = os.path.join(args.result_dir, input_basename)
+            os.makedirs(save_dir_full,exist_ok = True)
+            cmd = f"ffmpeg -v fatal -i {video_path} -start_number 0 {save_dir_full}/%08d.png"
+            os.system(cmd)
+            input_img_list = sorted(glob.glob(os.path.join(save_dir_full, '*.[jpJP][pnPN]*[gG]')))
+            fps = get_video_fps(video_path)
+        elif get_file_type(video_path)=="image":
+            input_img_list = [video_path, ]
+            fps = args.fps
+        elif os.path.isdir(video_path):  # input img folder
+            input_img_list = glob.glob(os.path.join(video_path, '*.[jpJP][pnPN]*[gG]'))
+            input_img_list = sorted(input_img_list, key=lambda x: int(os.path.splitext(os.path.basename(x))[0]))
+            fps = args.fps
+        else:
+            raise ValueError(f"{video_path} should be a video file, an image file or a directory of images")
+        ############################################## extract audio feature ##############################################
+        whisper_feature = audio_processor.audio2feat(audio_path)
+        for __ in range(0, len(whisper_feature) - 1, 2):  # -1 to avoid index error if the list has an odd number of elements
+            # Combine two consecutive chunks
+            # pair_of_chunks = np.array([whisper_feature[__], whisper_feature[__+1]])
+            concatenated_chunks = np.concatenate([whisper_feature[__], whisper_feature[__+1]], axis=0)
+            # Save the pair to a .npy file
+            np.save(f'data/audios/{folder_name}/{total_audio_index+(__//2)+1}.npy', concatenated_chunks)
+            temp_audio_index=(__//2)+total_audio_index+1
+        total_audio_index=temp_audio_index
+        whisper_chunks = audio_processor.feature2chunks(feature_array=whisper_feature,fps=fps)
+
+        ############################################## preprocess input image  ##############################################
+        gc.collect()
+        if os.path.exists(crop_coord_save_path) and args.use_saved_coord:
+            print("using extracted coordinates")
+            with open(crop_coord_save_path,'rb') as f:
+                coord_list = pickle.load(f)
+            frame_list = read_imgs(input_img_list)
+        else:
+            print("extracting landmarks...time consuming")
+            coord_list, frame_list = get_landmark_and_bbox(input_img_list, bbox_shift)
+            with open(crop_coord_save_path, 'wb') as f:
+                pickle.dump(coord_list, f)
+
+                
+        i = 0
+        input_latent_list = []
+        crop_i=0
+        crop_data=[]
+        for bbox, frame in zip(coord_list, frame_list):
+            if bbox == coord_placeholder:
+                continue
+            x1, y1, x2, y2 = bbox
+
+            x1=max(0,x1)
+            y1=max(0,y1)
+            x2=max(0,x2)
+            y2=max(0,y2)
+
+            if ((y2-y1)<=0) or ((x2-x1)<=0):
+                continue
+            crop_frame = frame[y1:y2, x1:x2]
+            crop_frame = cv2.resize(crop_frame,(256,256),interpolation = cv2.INTER_LANCZOS4)
+            latents = vae.get_latents_for_unet(crop_frame)
+            crop_data.append(crop_frame)
+            input_latent_list.append(latents)
+            crop_i+=1
+    
+        # to smooth the first and the last frame
+        frame_list_cycle = frame_list + frame_list[::-1]
+        coord_list_cycle = coord_list + coord_list[::-1]
+        input_latent_list_cycle = input_latent_list + input_latent_list[::-1]
+        crop_data = crop_data + crop_data[::-1]
+        ############################################## inference batch by batch ##############################################
+        video_num = len(whisper_chunks)
+        batch_size = args.batch_size
+        gen = datagen(whisper_chunks,crop_data,batch_size)
+        crop_index = 0
+        for i, (whisper_batch,crop_batch) in enumerate(tqdm(gen,total=int(np.ceil(float(video_num)/batch_size)))):
+            for image,audio in zip(crop_batch,whisper_batch):
+                cv2.imwrite(f"data/images/{folder_name}/{str(crop_index+total_image_index+1)}.png",image)
+                temp_image_index = crop_index + total_image_index + 1
+                crop_index += 1
+                
+                # np.save(f'data/audios/{folder_name}/{str(i+crop_index)}.npy', audio)
+        total_image_index=temp_image_index
+        gc.collect()
+
+           
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--inference_config", type=str, default="configs/inference/test_img.yaml")
+    parser.add_argument("--bbox_shift", type=int, default=0)
+    parser.add_argument("--result_dir", default='./results', help="path to output")
+    parser.add_argument("--folder_name", default=f'{uuid.uuid4()}', help="path to output")
+
+    parser.add_argument("--fps", type=int, default=25)
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--output_vid_name", type=str, default=None)
+    parser.add_argument("--use_saved_coord",
+                        action="store_true",
+                        help='use saved coordinate to save time')
+    parser.add_argument("--use_float16",
+                        action="store_true",
+                        help="Whether use float16 to speed up inference",
+    )
+
+    args = parser.parse_args()
+    main(args)
+
+
+def process_audio(audio_path):
+    whisper_feature = audio_processor.audio2feat(audio_path)
+    np.save('audio/your_filename.npy', whisper_feature)
+
+def mask_face(image):
+    # Load dlib's face detector and the landmark predictor
+    detector = dlib.get_frontal_face_detector()
+    predictor_path = "/content/shape_predictor_68_face_landmarks.dat"  # Set path to your downloaded predictor file
+    predictor = dlib.shape_predictor(predictor_path)
+
+    # Load your input image
+    # image_path = "/content/ori_frame_00000077.png"  # Replace with the path to your input image
+    # image = cv2.imread(image_path)
+    if image is None:
+        raise ValueError("Image not found or unable to load.")
+
+    # Convert to grayscale for detection
+    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+
+    # Detect faces in the image
+    faces = detector(gray)
+
+    # Process each detected face
+    for face in faces:
+        # Predict landmarks
+        landmarks = predictor(gray, face)
+
+        # The indices of nose landmarks are 27 to 35
+        nose_tip = landmarks.part(33).y
+
+        # Blacken the region below the nose tip
+        blacken_area = image[nose_tip:, :]
+        blacken_area[:] = (0, 0, 0)
+
+    # Save the final image or display it
+    # cv2.imwrite("output_image.jpg", image)
+    return image

scripts/finetuned_inference.py

@@ -0,0 +1,182 @@
+import argparse
+import os
+from omegaconf import OmegaConf
+import numpy as np
+import cv2
+import torch
+import glob
+import pickle
+from tqdm import tqdm
+import copy
+
+from musetalk.utils.utils import get_file_type,get_video_fps,datagen
+from musetalk.utils.preprocessing import get_landmark_and_bbox,read_imgs,coord_placeholder
+from musetalk.utils.blending import get_image
+from musetalk.utils.utils import load_all_model
+import shutil
+
+from accelerate import Accelerator
+
+# load model weights
+audio_processor, vae, unet, pe = load_all_model()
+accelerator = Accelerator(
+        mixed_precision="fp16",
+    )
+unet = accelerator.prepare(
+        unet,
+
+    )
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+timesteps = torch.tensor([0], device=device)
+
+@torch.no_grad()
+def main(args):
+    global pe
+    if not (args.unet_checkpoint == None):
+        print("unet ckpt loaded")
+        accelerator.load_state(args.unet_checkpoint)
+
+    if args.use_float16 is True:
+        pe = pe.half()
+        vae.vae = vae.vae.half()
+        unet.model = unet.model.half()
+    
+    inference_config = OmegaConf.load(args.inference_config)
+    print(inference_config)
+    for task_id in inference_config:
+        video_path = inference_config[task_id]["video_path"]
+        audio_path = inference_config[task_id]["audio_path"]
+        bbox_shift = inference_config[task_id].get("bbox_shift", args.bbox_shift)
+
+        input_basename = os.path.basename(video_path).split('.')[0]
+        audio_basename  = os.path.basename(audio_path).split('.')[0]
+        output_basename = f"{input_basename}_{audio_basename}"
+        result_img_save_path = os.path.join(args.result_dir, output_basename) # related to video & audio inputs
+        crop_coord_save_path = os.path.join(result_img_save_path, input_basename+".pkl") # only related to video input
+        os.makedirs(result_img_save_path,exist_ok =True)
+        
+        if args.output_vid_name is None:
+            output_vid_name = os.path.join(args.result_dir, output_basename+".mp4")
+        else:
+            output_vid_name = os.path.join(args.result_dir, args.output_vid_name)
+        ############################################## extract frames from source video ##############################################
+        if get_file_type(video_path)=="video":
+            save_dir_full = os.path.join(args.result_dir, input_basename)
+            os.makedirs(save_dir_full,exist_ok = True)
+            cmd = f"ffmpeg -v fatal -i {video_path} -start_number 0 {save_dir_full}/%08d.png"
+            os.system(cmd)
+            input_img_list = sorted(glob.glob(os.path.join(save_dir_full, '*.[jpJP][pnPN]*[gG]')))
+            fps = get_video_fps(video_path)
+        elif get_file_type(video_path)=="image":
+            input_img_list = [video_path, ]
+            fps = args.fps
+        elif os.path.isdir(video_path):  # input img folder
+            input_img_list = glob.glob(os.path.join(video_path, '*.[jpJP][pnPN]*[gG]'))
+            input_img_list = sorted(input_img_list, key=lambda x: int(os.path.splitext(os.path.basename(x))[0]))
+            fps = args.fps
+        else:
+            raise ValueError(f"{video_path} should be a video file, an image file or a directory of images")
+        ############################################## extract audio feature ##############################################
+        whisper_feature = audio_processor.audio2feat(audio_path)
+        whisper_chunks = audio_processor.feature2chunks(feature_array=whisper_feature,fps=fps)
+        ############################################## preprocess input image  ##############################################
+        if os.path.exists(crop_coord_save_path) and args.use_saved_coord:
+            print("using extracted coordinates")
+            with open(crop_coord_save_path,'rb') as f:
+                coord_list = pickle.load(f)
+            frame_list = read_imgs(input_img_list)
+        else:
+            print("extracting landmarks...time consuming")
+            coord_list, frame_list = get_landmark_and_bbox(input_img_list, bbox_shift)
+            with open(crop_coord_save_path, 'wb') as f:
+                pickle.dump(coord_list, f)
+        
+                
+        i = 0
+        input_latent_list = []
+        crop_i=0
+        for bbox, frame in zip(coord_list, frame_list):
+            if bbox == coord_placeholder:
+                continue
+            x1, y1, x2, y2 = bbox
+            crop_frame = frame[y1:y2, x1:x2]
+            crop_frame = cv2.resize(crop_frame,(256,256),interpolation = cv2.INTER_LANCZOS4)
+            cv2.imwrite(f"{result_img_save_path}/crop_frame_{str(crop_i).zfill(8)}.png",crop_frame)
+            latents = vae.get_latents_for_unet(crop_frame)
+            input_latent_list.append(latents)
+            crop_i+=1
+    
+        # to smooth the first and the last frame
+        frame_list_cycle = frame_list + frame_list[::-1]
+        coord_list_cycle = coord_list + coord_list[::-1]
+        input_latent_list_cycle = input_latent_list + input_latent_list[::-1]
+        ############################################## inference batch by batch ##############################################
+        video_num = len(whisper_chunks)
+        batch_size = args.batch_size
+        gen = datagen(whisper_chunks,input_latent_list_cycle,batch_size)
+        res_frame_list = []
+        for i, (whisper_batch,latent_batch) in enumerate(tqdm(gen,total=int(np.ceil(float(video_num)/batch_size)))):
+            audio_feature_batch = torch.from_numpy(whisper_batch)
+            audio_feature_batch = audio_feature_batch.to(device=unet.device,
+                                                         dtype=unet.model.dtype) # torch, B, 5*N,384
+            audio_feature_batch = pe(audio_feature_batch)
+            latent_batch = latent_batch.to(dtype=unet.model.dtype)
+            pred_latents = unet.model(latent_batch, timesteps, encoder_hidden_states=audio_feature_batch).sample
+            recon = vae.decode_latents(pred_latents)
+            for res_frame in recon:
+                res_frame_list.append(res_frame)
+                
+        ############################################## pad to full image ##############################################
+        print("pad talking image to original video")
+        for i, res_frame in enumerate(tqdm(res_frame_list)):
+            bbox = coord_list_cycle[i%(len(coord_list_cycle))]
+            ori_frame = copy.deepcopy(frame_list_cycle[i%(len(frame_list_cycle))])
+            x1, y1, x2, y2 = bbox
+            try:
+                res_frame = cv2.resize(res_frame.astype(np.uint8),(x2-x1,y2-y1))
+            except:
+                continue
+            
+            combine_frame = get_image(ori_frame,res_frame,bbox)
+            cv2.imwrite(f"{result_img_save_path}/res_frame_{str(i).zfill(8)}.png",res_frame)
+            cv2.imwrite(f"{result_img_save_path}/ori_frame_{str(i).zfill(8)}.png",ori_frame)
+            cv2.imwrite(f"{result_img_save_path}/{str(i).zfill(8)}.png",combine_frame)
+
+        cmd_img2video = f"ffmpeg -y -v warning -r {fps} -f image2 -i {result_img_save_path}/%08d.png -vcodec libx264 -vf format=rgb24,scale=out_color_matrix=bt709,format=yuv420p -crf 18 temp.mp4"
+        os.system(cmd_img2video)
+        
+        cmd_combine_audio = f"ffmpeg -y -v warning -i {audio_path} -i temp.mp4 {output_vid_name}"
+        os.system(cmd_combine_audio)
+        
+        os.remove("temp.mp4")
+
+        cmd_img2video = f"ffmpeg -y -v warning -r {fps} -f image2 -i {result_img_save_path}/ori_frame_%08d.png -vcodec libx264 -vf format=rgb24,scale=out_color_matrix=bt709,format=yuv420p -crf 18 temp.mp4"
+        os.system(cmd_img2video)
+        
+        # cmd_combine_audio = f"ffmpeg -y -v warning -i {audio_path} -i temp.mp4 {output_vid_name}"
+        # print(cmd_combine_audio)
+        # os.system(cmd_combine_audio)
+
+        # shutil.rmtree(result_img_save_path)
+        print(f"result is save to {output_vid_name}")
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--inference_config", type=str, default="configs/inference/test_img.yaml")
+    parser.add_argument("--bbox_shift", type=int, default=0)
+    parser.add_argument("--result_dir", default='./results', help="path to output")
+
+    parser.add_argument("--fps", type=int, default=25)
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--output_vid_name", type=str, default=None)
+    parser.add_argument("--use_saved_coord",
+                        action="store_true",
+                        help='use saved coordinate to save time')
+    parser.add_argument("--use_float16",
+                        action="store_true",
+                        help="Whether use float16 to speed up inference",
+    )
+    parser.add_argument("--unet_checkpoint", type=str, default=None)
+
+    args = parser.parse_args()
+    main(args)

scripts/inference.py

@@ -158,4 +158,4 @@ if __name__ == "__main__":
     )
 
     args = parser.parse_args()
-    main(args)
+    main(args)

train_codes/DataLoader.py

@@ -0,0 +1,235 @@
+import os, random, cv2, argparse
+import torch
+from torch.utils import data as data_utils
+from os.path import dirname, join, basename, isfile
+import numpy as np
+from glob import glob
+from utils.utils import prepare_mask_and_masked_image
+import torchvision.utils as vutils
+import torchvision.transforms as transforms
+import shutil
+from tqdm import tqdm
+import ast
+import json
+import re
+import heapq
+
+syncnet_T = 1
+RESIZED_IMG = 256
+
+connections = [(0, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7),(7, 8), (8, 9), (9, 10), (10, 11), (11, 12), (12, 13),(13,14),(14,15),(15,16),  # 下颌线
+                       (17, 18), (18, 19), (19, 20), (20, 21), #左眉毛
+                       (22, 23), (23, 24), (24, 25), (25, 26), #右眉毛
+                       (27, 28),(28,29),(29,30),# 鼻梁
+                       (31,32),(32,33),(33,34),(34,35), #鼻子
+                       (36,37),(37,38),(38, 39), (39, 40), (40, 41), (41, 36), # 左眼
+                       (42, 43), (43, 44), (44, 45), (45, 46), (46, 47), (47, 42), # 右眼
+                       (48, 49),(49, 50), (50, 51),(51, 52),(52, 53), (53, 54), # 上嘴唇 外延
+                       (54, 55), (55, 56), (56, 57), (57, 58), (58, 59), (59, 48),  # 下嘴唇 外延
+                       (60, 61), (61, 62), (62, 63), (63, 64), (64, 65),  (65, 66), (66, 67), (67, 60) #嘴唇内圈
+              ]  
+
+
+def get_image_list(data_root, split):
+    filelist = []
+    imgNumList = []
+    with open('filelists/{}.txt'.format(split)) as f:
+        for line in f:
+            line = line.strip()
+            if ' ' in line:
+                filename = line.split()[0]
+                imgNum = int(line.split()[1])
+                filelist.append(os.path.join(data_root, filename))
+            imgNumList.append(imgNum)
+    return filelist, imgNumList
+
+
+
+class Dataset(object):
+    def __init__(self, 
+                 data_root, 
+                 json_path, 
+                 use_audio_length_left=1,
+                 use_audio_length_right=1,
+                 whisper_model_type = "tiny"
+                 ):
+        # self.all_videos, self.all_imgNum = get_image_list(data_root, split)
+        self.audio_feature = [use_audio_length_left,use_audio_length_right]
+        self.all_img_names = []
+        # self.split = split
+        self.img_names_path = '../data'
+        self.whisper_model_type = whisper_model_type
+        self.use_audio_length_left = use_audio_length_left
+        self.use_audio_length_right = use_audio_length_right
+
+        if self.whisper_model_type =="tiny":
+            self.whisper_path = '../data/audios'
+            self.whisper_feature_W = 5
+            self.whisper_feature_H = 384
+        elif self.whisper_model_type =="largeV2":
+            self.whisper_path = '...'
+            self.whisper_feature_W = 33
+            self.whisper_feature_H = 1280
+        self.whisper_feature_concateW = self.whisper_feature_W*2*(self.use_audio_length_left+self.use_audio_length_right+1) #5*2*（2+2+1）= 50
+        with open(json_path, 'r') as file:
+            self.all_videos = json.load(file)
+
+        for vidname in tqdm(self.all_videos, desc="Preparing dataset"):
+            json_path_names = f"{self.img_names_path}/{vidname.split('/')[-1].split('.')[0]}.json"
+            if not os.path.exists(json_path_names):
+                img_names = glob(join(vidname, '*.png'))
+                img_names.sort(key=lambda x:int(x.split("/")[-1].split('.')[0]))
+                with open(json_path_names, "w") as f:
+                    json.dump(img_names,f)
+            else:
+                with open(json_path_names, "r") as f:
+                    img_names = json.load(f)
+            self.all_img_names.append(img_names)
+            
+    def get_frame_id(self, frame):
+        return int(basename(frame).split('.')[0])
+
+    def get_window(self, start_frame):
+        start_id = self.get_frame_id(start_frame)
+        vidname = dirname(start_frame)
+
+        window_fnames = []
+        for frame_id in range(start_id, start_id + syncnet_T):
+            frame = join(vidname, '{}.png'.format(frame_id))
+            if not isfile(frame):
+                return None
+            window_fnames.append(frame)
+        return window_fnames
+    
+    def read_window(self, window_fnames):
+        if window_fnames is None: return None
+        window = []
+        for fname in window_fnames:
+            img = cv2.imread(fname)
+            if img is None:
+                return None
+            try:
+                img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+                img = cv2.resize(img, (RESIZED_IMG, RESIZED_IMG))
+            except Exception as e:
+                print("read_window has error fname not exist:",fname)
+                return None
+
+            window.append(img)
+
+        return window
+    
+    def prepare_window(self, window):
+        #  1 x H x W x 3
+        x = np.asarray(window) / 255.
+        x = np.transpose(x, (3, 0, 1, 2))
+
+        return x
+
+    def __len__(self):
+        return len(self.all_videos)
+
+    def __getitem__(self, idx):
+        while 1:
+            idx = random.randint(0, len(self.all_videos) - 1)
+            #随机选择某个video里
+            vidname = self.all_videos[idx].split('/')[-1]
+            video_imgs = self.all_img_names[idx]
+            if len(video_imgs) <= 11:
+                continue
+            img_name = random.choice(video_imgs)
+            img_idx = int(basename(img_name).split(".")[0])
+            random_element = random.randint(0,len(video_imgs)-1)
+            while abs(random_element - img_idx) <= 5:
+                random_element = random.randint(0,len(video_imgs)-1)
+            img_dir = os.path.dirname(img_name)
+            ref_image = os.path.join(img_dir, f"{str(random_element)}.png")
+            target_window_fnames = self.get_window(img_name)
+            ref_window_fnames = self.get_window(ref_image)
+           
+            if target_window_fnames is None or ref_window_fnames is None:
+                print("No such img",img_name, ref_image)
+                continue
+                 
+            try:
+                #构建目标img数据
+                target_window = self.read_window(target_window_fnames)
+                if target_window is None :
+                    print("No such target window,",target_window_fnames)
+                    continue
+                #构建参考img数据
+                ref_window = self.read_window(ref_window_fnames)
+    
+                if ref_window is None:
+                    print("No such target ref window,",ref_window)
+                    continue
+            except Exception as e:
+                print(f"发生未知错误：{e}")
+                continue
+          
+            #构建target输入
+            target_window = self.prepare_window(target_window)
+            image = gt = target_window.copy().squeeze()
+            target_window[:, :, target_window.shape[2]//2:] = 0.                   # upper half face, mask掉下半部分        V1：输入       
+            ref_image = self.prepare_window(ref_window).squeeze()   
+       
+            
+
+            mask = torch.zeros((ref_image.shape[1], ref_image.shape[2]))
+            mask[:ref_image.shape[2]//2,:] = 1
+            image = torch.FloatTensor(image)
+            mask, masked_image = prepare_mask_and_masked_image(image,mask)
+           
+            
+            
+            #音频特征
+            window_index = self.get_frame_id(img_name)
+            sub_folder_name = vidname.split('/')[-1]
+            
+            ## 根据window_index加载相邻的音频
+            audio_feature_all = []
+            is_index_out_of_range = False
+            if os.path.isdir(os.path.join(self.whisper_path, sub_folder_name)):
+                for feat_idx in range(window_index-self.use_audio_length_left,window_index+self.use_audio_length_right+1):
+                    # 判定是否越界
+                    audio_feat_path = os.path.join(self.whisper_path, sub_folder_name, str(feat_idx) + ".npy")
+                    if not os.path.exists(audio_feat_path):
+                        is_index_out_of_range = True
+                        break
+
+                    try:
+                        audio_feature_all.append(np.load(audio_feat_path))
+                    except Exception as e:
+                        print(f"发生未知错误：{e}")
+                        print(f"npy load error {audio_feat_path}")
+                if is_index_out_of_range:
+                    continue
+                audio_feature = np.concatenate(audio_feature_all, axis=0)
+            else:
+                continue
+
+            audio_feature = audio_feature.reshape(1, -1, self.whisper_feature_H) #1， -1， 384
+            if audio_feature.shape != (1,self.whisper_feature_concateW, self.whisper_feature_H):  #1 50 384
+                print(f"shape error!! {vidname} {window_index}, audio_feature.shape: {audio_feature.shape}")
+                continue
+            audio_feature = torch.squeeze(torch.FloatTensor(audio_feature))
+            return ref_image, image, masked_image, mask, audio_feature
+         
+    
+    
+if __name__ == "__main__":
+    data_root = '...'
+    val_data = Dataset(data_root, 
+                          'val', 
+                          use_audio_length_left = 2,
+                          use_audio_length_right = 2,
+                          whisper_model_type = "tiny"
+                          )
+    val_data_loader = data_utils.DataLoader(
+        val_data, batch_size=4, shuffle=True,
+        num_workers=1)
+
+    for i, data in enumerate(val_data_loader):
+        ref_image, image, masked_image, mask, audio_feature = data
+
+ 

train_codes/README.md

@@ -0,0 +1,52 @@
+# Data preprocessing
+
+Create two config yaml files, one for training and other for testing (both in same format as configs/inference/test.yaml)
+The train yaml file should contain the training video paths and corresponding audio paths
+The test yaml file should contain the validation video paths and corresponding audio paths
+
+Run:
+```
+./data_new.sh train output train_video1.mp4 train_video2.mp4
+./data_new.sh test output test_video1.mp4 test_video2.mp4
+```
+This creates folders which contain the image frames and npy files. This also creates train.json and val.json which can be used during the training.
+
+## Data organization
+```
+./data/
+├── images
+│     └──RD_Radio10_000
+│         └── 0.png
+│         └── 1.png
+│         └── xxx.png
+│     └──RD_Radio11_000
+│         └── 0.png
+│         └── 1.png
+│         └── xxx.png
+├── audios
+│     └──RD_Radio10_000
+│         └── 0.npy
+│         └── 1.npy
+│         └── xxx.npy
+│     └──RD_Radio11_000
+│         └── 0.npy
+│         └── 1.npy
+│         └── xxx.npy
+```
+
+## Training
+Simply run after preparing the preprocessed data
+```
+cd train_codes
+sh train.sh #--train_json="../train.json" \(Generated in Data preprocessing step.)
+            #--val_json="../val.json" \
+```
+## Inference with trained checkpoit
+Simply run after training the model, the model checkpoints are saved at train_codes/output usually
+```
+python -m scripts.finetuned_inference --inference_config configs/inference/test.yaml --unet_checkpoint path_to_trained_checkpoint_folder
+```
+
+## TODO
+- [x] release data preprocessing codes
+- [ ] release some novel designs in training (after technical report)

train_codes/filelists/train.txt

@@ -0,0 +1,322 @@
+RD_Radio10_000 3501
+RD_Radio11_000 752
+RD_Radio11_001 1502
+RD_Radio12_000 876
+RD_Radio13_000 877
+RD_Radio14_000 1251
+RD_Radio16_000 2377
+RD_Radio17_000 2176
+RD_Radio18_000 1827
+RD_Radio19_000 1876
+RD_Radio1_000 1876
+RD_Radio20_000 276
+RD_Radio21_000 1201
+RD_Radio22_000 752
+RD_Radio23_000 1602
+RD_Radio25_000 2126
+RD_Radio26_000 1052
+RD_Radio27_000 1376
+RD_Radio28_000 2252
+RD_Radio29_000 2252
+RD_Radio2_000 2076
+RD_Radio30_000 2877
+RD_Radio31_000 1377
+RD_Radio32_000 1502
+RD_Radio33_000 2252
+RD_Radio34_000 702
+RD_Radio34_001 252
+RD_Radio34_002 501
+RD_Radio34_003 326
+RD_Radio34_004 502
+RD_Radio34_005 502
+RD_Radio34_006 502
+RD_Radio34_007 252
+RD_Radio34_008 876
+RD_Radio34_009 752
+RD_Radio35_000 2127
+RD_Radio36_000 2377
+RD_Radio37_000 3127
+RD_Radio38_000 3752
+RD_Radio39_000 1502
+RD_Radio3_000 2377
+RD_Radio40_000 1252
+RD_Radio41_000 2127
+RD_Radio42_000 1752
+RD_Radio43_000 1502
+RD_Radio44_000 1127
+RD_Radio45_000 1628
+RD_Radio46_000 1877
+RD_Radio47_000 1502
+RD_Radio48_000 2127
+RD_Radio49_000 1502
+RD_Radio4_000 1002
+RD_Radio50_000 2252
+RD_Radio51_000 3253
+RD_Radio52_000 2752
+RD_Radio53_000 2627
+RD_Radio54_000 577
+RD_Radio56_000 1952
+RD_Radio57_000 3077
+RD_Radio59_000 1952
+RD_Radio5_000 1377
+RD_Radio7_000 2252
+RD_Radio8_000 2126
+RD_Radio9_000 1502
+WDA_AdamSchiff_000 6877
+WDA_AdamSmith_000 7327
+WDA_AlexandriaOcasioCortez_000 2252
+WDA_AmyKlobuchar0_000 6501
+WDA_AmyKlobuchar1_000 3253
+WDA_AmyKlobuchar1_001 877
+WDA_AmyKlobuchar1_002 1502
+WDA_AmyKlobuchar1_003 1377
+WDA_AndyKim_000 6952
+WDA_AndyLevin_000 4876
+WDA_AnnieKuster_000 4876
+WDA_BarackObama_000 3575
+WDA_BarackObama_001 5625
+WDA_BarbaraLee0_000 6502
+WDA_BarbaraLee1_000 4702
+WDA_BenCardin0_000 8127
+WDA_BenCardin1_000 7102
+WDA_BenRayLujn_000 7127
+WDA_BennieThompson1_000 6375
+WDA_BennieThompson_000 5375
+WDA_BernieSanders_000 8451
+WDA_BettyMcCollum_000 7002
+WDA_BillPascrell_000 7252
+WDA_BillRichardson_000 3127
+WDA_BobCasey0_000 10627
+WDA_BobCasey1_000 252
+WDA_BobMenendez_000 3002
+WDA_BobbyScott_000 2002
+WDA_BradSchneider_000 6627
+WDA_BrendaLawrence_000 5627
+WDA_BrianSchatz0_000 502
+WDA_BrianSchatz1_000 2627
+WDA_BrianSchatz2_000 1952
+WDA_ByronDorgan1_000 4277
+WDA_CarolynMaloney1_000 8377
+WDA_CatherineCortezMasto_000 2827
+WDA_CedricRichmond_000 7250
+WDA_ChrisCoons1_000 4452
+WDA_ChrisCoons_000 6827
+WDA_ChrisMurphy0_000 6877
+WDA_ChrisMurphy1_000 6377
+WDA_ChrisVanHollen0_000 8202
+WDA_ChrisVanHollen1_000 7327
+WDA_ChuckSchumer0_000 5577
+WDA_ChuckSchumer1_000 4827
+WDA_ColinAllred_000 4751
+WDA_DanKildee1_000 6252
+WDA_DanKildee_000 2502
+WDA_DavidCicilline_000 5875
+WDA_DebHaaland_000 5876
+WDA_DebbieDingell0_000 7752
+WDA_DebbieDingell1_000 7251
+WDA_DebbieStabenow0_000 4577
+WDA_DebbieStabenow1_000 5201
+WDA_DebbieWassermanSchultz_000 7625
+WDA_DianaDeGette0_000 6002
+WDA_DianaDeGette1_000 2202
+WDA_DianneFeinstein_000 7453
+WDA_DickDurbin_000 6326
+WDA_DonaldMcEachin_000 7627
+WDA_DonnaShalala1_000 7500
+WDA_DougJones_000 6077
+WDA_EdMarkey0_000 4752
+WDA_EdMarkey1_000 6501
+WDA_ElijahCummings_000 6377
+WDA_EliotEngel_000 6876
+WDA_EmanuelCleaver_000 7001
+WDA_EricSwalwell_000 6377
+WDA_FrankPallone0_000 5625
+WDA_FrankPallone1_000 6502
+WDA_GerryConnolly_000 6752
+WDA_HakeemJeffries_000 6002
+WDA_HaleyStevens_000 5001
+WDA_HenryWaxman_000 1125
+WDA_HillaryClinton_000 2500
+WDA_JackReed0_000 2377
+WDA_JackReed1_000 4877
+WDA_JackieSpeier_000 4625
+WDA_JackyRosen1_000 10077
+WDA_JackyRosen_000 5502
+WDA_JamesClyburn1_000 6876
+WDA_JamesClyburn_000 6875
+WDA_JanSchakowsky0_000 4128
+WDA_JanSchakowsky1_000 6251
+WDA_JeanneShaheen0_000 6702
+WDA_JeanneShaheen1_000 6577
+WDA_JeffMerkley1_000 6952
+WDA_JerryNadler_000 8377
+WDA_JimHimes_000 7000
+WDA_JimmyGomez_000 4627
+WDA_JoaquinCastro_000 5126
+WDA_JoeCrowley0_000 4877
+WDA_JoeCrowley1_000 1127
+WDA_JoeCrowley1_001 627
+WDA_JoeCrowley1_002 502
+WDA_JoeCrowley1_003 752
+WDA_JoeDonnelly_000 1377
+WDA_JoeKennedy_000 1327
+WDA_JoeManchin_000 3377
+WDA_JoeNeguse_000 1628
+WDA_JoeNeguse_001 1126
+WDA_JoeNeguse_002 1251
+WDA_JohnLewis0_000 6252
+WDA_JohnLewis1_000 7252
+WDA_JohnSarbanes0_000 6377
+WDA_JohnSarbanes1_000 1877
+WDA_JohnYarmuth1_000 5377
+WDA_JonTester0_000 3252
+WDA_JonTester1_000 5327
+WDA_KarenBass_000 7126
+WDA_KatherineClark_000 1552
+WDA_KathyCastor1_000 6001
+WDA_KathyCastor_000 2252
+WDA_KimSchrier_000 6877
+WDA_KirstenGillibrand_000 9627
+WDA_LaurenUnderwood_000 9877
+WDA_LisaBluntRochester_000 4500
+WDA_LloydDoggett0_000 7377
+WDA_LloydDoggett1_000 2252
+WDA_LucilleRoybal-Allard_000 2877
+WDA_LucyMcBath_000 4000
+WDA_MarciaFudge_000 7377
+WDA_MarkWarner1_000 377
+WDA_MarkWarner1_001 1327
+WDA_MarkWarner2_000 377
+WDA_MarkWarner_000 3127
+WDA_MartinHeinrich_000 5202
+WDA_MattCartwright_000 5377
+WDA_MazieHirono0_000 3752
+WDA_MichelleLujanGrisham_000 6875
+WDA_MichelleObama_000 2000
+WDA_MikeDoyle_000 8750
+WDA_MikeThompson0_000 4625
+WDA_MikeThompson1_000 1827
+WDA_NancyPelosi0_000 10251
+WDA_NancyPelosi1_000 1377
+WDA_NancyPelosi3_000 1127
+WDA_NitaLowey_000 5876
+WDA_NydiaVelzquez_000 5500
+WDA_PatrickLeahy0_000 6951
+WDA_PatrickLeahy1_000 9953
+WDA_PattyMurray0_000 3627
+WDA_PattyMurray1_000 5702
+WDA_PeterDeFazio_000 6628
+WDA_RaulRuiz_000 5752
+WDA_RichardBlumenthal_000 7827
+WDA_RichardNeal0_000 6252
+WDA_RichardNeal1_000 6877
+WDA_RobinKelly_000 4377
+WDA_RonWyden0_000 5152
+WDA_RonWyden1_000 8327
+WDA_ScottPeters0_000 7002
+WDA_ScottPeters1_000 3952
+WDA_SeanCasten_000 6126
+WDA_SeanPatrickMaloney_000 6502
+WDA_SheldonWhitehouse0_000 6327
+WDA_SheldonWhitehouse1_000 5702
+WDA_SherrodBrown0_000 7452
+WDA_SherrodBrown1_000 7327
+WDA_StenyHoyer_000 3502
+WDA_StephanieMurphy_000 4000
+WDA_SuzanDelBene_000 6875
+WDA_TammyBaldwin0_000 5327
+WDA_TammyBaldwin1_000 2503
+WDA_TammyDuckworth_000 5702
+WDA_TedLieu_000 6877
+WDA_TerriSewell0_000 2127
+WDA_TerriSewell1_000 10952
+WDA_TerriSewell_000 6752
+WDA_TimWalz_000 6628
+WDA_TinaSmith_000 5576
+WDA_TomCarper_000 4701
+WDA_TomPerez_000 5500
+WDA_TomUdall_000 4077
+WDA_VeronicaEscobar0_000 4377
+WDA_VeronicaEscobar1_000 2453
+WDA_WhipJimClyburn_000 6876
+WDA_XavierBecerra_000 877
+WDA_XavierBecerra_001 627
+WDA_XavierBecerra_002 1377
+WDA_ZoeLofgren_000 4625
+WRA_AdamKinzinger0_000 4251
+WRA_AdamKinzinger1_000 2751
+WRA_AdamKinzinger2_000 1002
+WRA_AdamKinzinger2_001 1001
+WRA_AdamKinzinger2_002 502
+WRA_AllenWest_000 6876
+WRA_AnnMarieBuerkle_000 452
+WRA_AnnWagner_000 3127
+WRA_AustinScott0_000 1177
+WRA_BillCassidy0_000 427
+WRA_BobCorker_000 1127
+WRA_BobGoodlatte0_000 1001
+WRA_BobGoodlatte0_001 752
+WRA_BobGoodlatte0_002 876
+WRA_BobGoodlatte0_003 1002
+WRA_BobbySchilling_001 1000
+WRA_BobbySchilling_002 1001
+WRA_CandiceMiller0_000 3001
+WRA_CarlyFiorina0_000 876
+WRA_CarlyFiorina_000 1902
+WRA_CathyMcMorrisRodgers0_000 6077
+WRA_CathyMcMorrisRodgers1_000 7375
+WRA_CathyMcMorrisRodgers1_001 7500
+WRA_CathyMcMorrisRodgers1_002 7500
+WRA_CathyMcMorrisRodgers2_000 2751
+WRA_ChuckGrassley_000 4502
+WRA_CoryGardner0_000 8577
+WRA_CoryGardner1_000 2002
+WRA_CoryGardner_000 1252
+WRA_DanNewhouse_000 3627
+WRA_DanSullivan_000 6877
+WRA_DaveCamp_000 752
+WRA_DaveCamp_001 876
+WRA_DaveCamp_002 627
+WRA_DavidVitter_000 2877
+WRA_DeanHeller_000 377
+WRA_DebFischer0_000 7577
+WRA_DebFischer1_000 1577
+WRA_DebFischer2_000 5202
+WRA_DianeBlack0_000 451
+WRA_DianeBlack0_001 301
+WRA_DianeBlack1_000 250
+WRA_DuncanHunter_000 727
+WRA_EricCantor_000 2952
+WRA_ErikPaulsen_000 876
+WRA_ErikPaulsen_001 377
+WRA_ErikPaulsen_002 627
+WRA_ErikPaulsen_003 752
+WRA_FredUpton_000 1701
+WRA_GeoffDavis_000 250
+WRA_GeorgeLeMieux_000 752
+WRA_GeorgeLeMieux_001 1001
+WRA_GregWalden1_000 752
+WRA_GregWalden1_001 1127
+WRA_GregWalden1_002 1327
+WRA_GregWalden_000 377
+WRA_JaimeHerreraBeutler0_000 952
+WRA_JebHensarling0_001 1176
+WRA_JebHensarling1_000 2327
+WRA_JebHensarling1_001 727
+WRA_JebHensarling2_000 2302
+WRA_JebHensarling2_001 877
+WRA_JebHensarling2_002 2127
+WRA_JebHensarling2_003 1877
+WRA_JeffFlake_000 7202
+WRA_JeffFlake_001 7502
+WRA_JeffFlake_002 7377
+WRA_JimInhofe_000 3127
+WRA_JimRisch_000 1377
+WRA_JoeHeck1_000 250
+WRA_JoePitts_000 1077
+WRA_JohnBarrasso0_000 5077
+WRA_JohnBarrasso1_000 3452
+WRA_JohnBoehner0_000 1626
+WRA_JohnBoehner1_000 2951
+WRA_JohnBoozman_000 1877
+WRA_JohnHoeven_000 2577

train_codes/filelists/val.txt

@@ -0,0 +1,30 @@
+WRA_JohnKasich0_000 1302
+WRA_JohnKasich1_000 1301
+WRA_JohnKasich1_001 1127
+WRA_JohnKasich3_000 1052
+WRA_JohnThune_000 1452
+WRA_JohnnyIsakson_000 5951
+WRA_JohnnyIsakson_001 5000
+WRA_JonKyl_000 626
+WRA_JoniErnst0_000 2077
+WRA_JoniErnst1_000 1452
+WRA_JuddGregg_000 1252
+WRA_JuddGregg_001 952
+WRA_JuddGregg_002 753
+WRA_KayBaileyHutchison_000 6325
+WRA_KellyAyotte_000 8077
+WRA_KevinBrady2_000 1276
+WRA_KevinBrady3_000 752
+WRA_KevinBrady_000 2503
+WRA_KevinMcCarthy0_000 1002
+WRA_KevinMcCarthy0_001 1127
+WRA_KristiNoem0_000 727
+WRA_KristiNoem1_000 452
+WRA_KristiNoem2_000 5952
+WRA_KristiNoem2_001 7277
+WRA_LamarAlexander0_000 1527
+WRA_LamarAlexander_000 1552
+WRA_LisaMurkowski0_000 1752
+WRA_LynnJenkins_000 877
+WRA_LynnJenkins_001 1002
+WRA_MarcoRubio_000 526

train_codes/musetalk.json

@@ -0,0 +1,36 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.6.0.dev0",
+  "act_fn": "silu",
+  "attention_head_dim": 8,
+  "block_out_channels": [
+    320,
+    640,
+    1280,
+    1280
+  ],
+  "center_input_sample": false,
+  "cross_attention_dim": 384,
+  "down_block_types": [
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "DownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 8,
+  "layers_per_block": 2,
+  "mid_block_scale_factor": 1,
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "out_channels": 4,
+  "sample_size": 64,
+  "up_block_types": [
+    "UpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D"
+  ]
+}

train_codes/train.py

@@ -0,0 +1,670 @@
+import argparse
+import itertools
+import math
+import os
+import random
+from pathlib import Path
+import json
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import ProjectConfiguration, set_seed
+from PIL import Image, ImageDraw
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    StableDiffusionInpaintPipeline,
+    StableDiffusionPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.optimization import get_scheduler
+from diffusers.utils import check_min_version
+
+import sys
+sys.path.append("./")
+
+from DataLoader import Dataset
+from utils.utils import preprocess_img_tensor
+from torch.utils import data as data_utils
+from utils.model_utils import validation,PositionalEncoding
+import time
+import pandas as pd
+from PIL import Image
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.13.0.dev0")
+
+logger = get_logger(__name__)
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--unet_config_file",
+        type=str,
+        default=None,
+        required=True,
+        help="the configuration of unet file.",
+    )
+    parser.add_argument(
+        "--reconstruction",
+        default=False,
+        action="store_true",
+        help="Flag to add prior preservation loss.",
+    )
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        required=True,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+   
+    parser.add_argument(
+        "--data_root",
+        type=str,
+        default=None,
+        required=True,
+        help="A folder containing the training data of instance images.",
+    )
+
+
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="text-inversion-model",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
+    parser.add_argument("--testing_speed", action="store_true", help="Whether to caculate the running time")
+    parser.add_argument(
+        "--train_batch_size", type=int, default=4, help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument("--num_train_epochs", type=int, default=1)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate",
+        type=float,
+        default=5e-6,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        action="store_true",
+        default=False,
+        help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
+    )
+    parser.add_argument(
+        "--lr_scheduler",
+        type=str,
+        default="constant",
+        help=(
+            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
+            ' "constant", "constant_with_warmup"]'
+        ),
+    )
+    parser.add_argument(
+        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
+    )
+    parser.add_argument(
+        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
+    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
+    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument("--train_json", type=str, default="train.json", help="The json file containing train image folders")
+    parser.add_argument("--val_json", type=str, default="test.json", help="The json file containing validation image folders")
+    parser.add_argument(
+        "--hub_model_id",
+        type=str,
+        default=None,
+        help="The name of the repository to keep in sync with the local `output_dir`.",
+    )
+    parser.add_argument(
+        "--logging_dir",
+        type=str,
+        default="logs",
+        help=(
+            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
+            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default="no",
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose"
+            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
+            "and an Nvidia Ampere GPU."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=1000,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints can be used both as final"
+            " checkpoints in case they are better than the last checkpoint and are suitable for resuming training"
+            " using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--validation_steps",
+        type=int,
+        default=1000,
+        help=(
+            "Conduct validation every X updates."
+        ),
+    )
+    parser.add_argument(
+        "--val_out_dir",
+        type=str,
+        default = '',
+        help=(
+            "Conduct validation every X updates."
+        ),
+    )
+    
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=(
+            "Max number of checkpoints to store. Passed as `total_limit` to the `Accelerator` `ProjectConfiguration`."
+            " See Accelerator::save_state https://huggingface.co/docs/accelerate/package_reference/accelerator#accelerate.Accelerator.save_state"
+            " for more docs"
+        ),
+    )
+    parser.add_argument(
+        "--resume_from_checkpoint",
+        type=str,
+        default=None,
+        help=(
+            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
+            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
+        ),
+    )
+    parser.add_argument(
+        "--use_audio_length_left",
+        type=int,
+        default=1,
+        help="number of audio length (left).",
+    )
+    parser.add_argument(
+        "--use_audio_length_right",
+        type=int,
+        default=1,
+        help="number of audio length (right)",
+    )
+    parser.add_argument(
+        "--whisper_model_type",
+        type=str,
+        default="landmark_nearest",
+        choices=["tiny","largeV2"],
+        help="Determine whisper feature type",
+    )
+
+    args = parser.parse_args()
+    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
+    if env_local_rank != -1 and env_local_rank != args.local_rank:
+        args.local_rank = env_local_rank
+    
+    return args
+
+def print_model_dtypes(model, model_name):
+    for name, param in model.named_parameters():
+        if(param.dtype!=torch.float32):
+            print(f"{name}: {param.dtype}")
+
+
+def main():
+    args = parse_args()
+    args.output_dir = f"output/{args.output_dir}"
+    args.val_out_dir = f"val/{args.val_out_dir}"
+    os.makedirs(args.output_dir, exist_ok=True)
+    os.makedirs(args.val_out_dir, exist_ok=True)
+    
+    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder='vae')
+    
+    
+    logging_dir = Path(args.output_dir, args.logging_dir)
+
+    project_config = ProjectConfiguration(
+        total_limit=args.checkpoints_total_limit, project_dir=args.output_dir, logging_dir=logging_dir
+    )
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with="tensorboard",
+        project_config=project_config,
+    )
+
+    # Currently, it's not possible to do gradient accumulation when training two models with accelerate.accumulate
+    # This will be enabled soon in accelerate. For now, we don't allow gradient accumulation when training two models.
+    # TODO (patil-suraj): Remove this check when gradient accumulation with two models is enabled in accelerate.
+    if args.gradient_accumulation_steps > 1 and accelerator.num_processes > 1:
+        raise ValueError(
+            "Gradient accumulation is not supported when training the text encoder in distributed training. "
+            "Please set gradient_accumulation_steps to 1. This feature will be supported in the future."
+        )
+
+    if args.seed is not None:
+#         set_seed(args.seed)
+        set_seed(seed + accelerator.process_index)
+
+
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+
+    # Load models and create wrapper for stable diffusion
+    with open(args.unet_config_file, 'r') as f:
+        unet_config = json.load(f)
+        
+    #text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder")
+    
+    # Todo:
+    print("Loading AutoencoderKL")
+    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder='vae')
+    vae_fp32 = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae")
+    print("Loading UNet2DConditionModel")
+    unet = UNet2DConditionModel(**unet_config)
+    
+    if args.whisper_model_type == "tiny":
+        pe = PositionalEncoding(d_model=384)
+    elif args.whisper_model_type == "largeV2":
+        pe = PositionalEncoding(d_model=1280)
+    else:
+        print(f"not support whisper_model_type {args.whisper_model_type}")
+        
+    print("Loading models done...")
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+
+    if args.scale_lr:
+        args.learning_rate = (
+            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
+    if args.use_8bit_adam:
+        try:
+            import bitsandbytes as bnb
+        except ImportError:
+            raise ImportError(
+                "To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`."
+            )
+
+        optimizer_class = bnb.optim.AdamW8bit
+    else:
+        optimizer_class = torch.optim.AdamW
+
+    params_to_optimize = (
+        itertools.chain(unet.parameters()))
+    optimizer = optimizer_class(
+        params_to_optimize,
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    print("loading train_dataset ...")
+    train_dataset = Dataset(args.data_root, 
+                            args.train_json, 
+                            use_audio_length_left=args.use_audio_length_left,
+                            use_audio_length_right=args.use_audio_length_right,
+                            whisper_model_type=args.whisper_model_type
+                            )
+    train_data_loader = data_utils.DataLoader(
+        train_dataset, batch_size=args.train_batch_size, shuffle=True,
+        num_workers=8)
+    print("loading val_dataset ...")
+    val_dataset = Dataset(args.data_root, 
+                          args.val_json, 
+                          use_audio_length_left=args.use_audio_length_left,
+                          use_audio_length_right=args.use_audio_length_right,
+                          whisper_model_type=args.whisper_model_type
+                         )
+    val_data_loader = data_utils.DataLoader(
+        val_dataset, batch_size=1, shuffle=False,
+        num_workers=8)
+
+    # Scheduler and math around the number of training steps.
+    overrode_max_train_steps = False
+    num_update_steps_per_epoch = math.ceil(len(train_data_loader) / args.gradient_accumulation_steps)
+    if args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        overrode_max_train_steps = True
+
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+    )
+
+  
+    
+    unet, optimizer, train_data_loader, val_data_loader, lr_scheduler = accelerator.prepare(
+        unet, optimizer, train_data_loader, val_data_loader,lr_scheduler
+    )
+    
+    vae.requires_grad_(False)
+    vae_fp32.requires_grad_(False)
+
+    weight_dtype = torch.float32
+    # weight_dtype = torch.float16
+    vae_fp32.to(accelerator.device, dtype=weight_dtype)
+    vae_fp32.encoder = None
+    if accelerator.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+    elif accelerator.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16    
+    vae.to(accelerator.device, dtype=weight_dtype)
+    vae.decoder = None
+    pe.to(accelerator.device, dtype=weight_dtype)
+
+    num_update_steps_per_epoch = math.ceil(len(train_data_loader) / args.gradient_accumulation_steps)
+    if overrode_max_train_steps:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+    # Afterwards we recalculate our number of training epochs
+    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+
+    # We need to initialize the trackers we use, and also store our configuration.
+    # The trackers initializes automatically on the main process.
+    if accelerator.is_main_process:
+        accelerator.init_trackers("dreambooth", config=vars(args))
+
+    # Train!
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+
+    print(f"  Num batches each epoch = {len(train_data_loader)}")
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Num examples = {len(train_dataset)}")
+    logger.info(f"  Num batches each epoch = {len(train_data_loader)}")
+    logger.info(f"  Num Epochs = {args.num_train_epochs}")
+    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
+    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
+    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
+    logger.info(f"  Total optimization steps = {args.max_train_steps}")
+    global_step = 0
+    first_epoch = 0
+
+    if args.resume_from_checkpoint:
+        if args.resume_from_checkpoint != "latest":
+            path = os.path.basename(args.resume_from_checkpoint)
+        else:
+            # Get the most recent checkpoint
+            dirs = os.listdir(args.output_dir)
+            dirs = [d for d in dirs if d.startswith("checkpoint")]
+            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
+            path = dirs[-1] if len(dirs) > 0 else None
+
+        # path="../models/pytorch_model.bin"
+        #TODO change path
+        # path=None
+        if path is None:
+            accelerator.print(
+                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            args.resume_from_checkpoint = None
+            print(f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run.")
+        else:
+            accelerator.print(f"Resuming from checkpoint {path}")
+            accelerator.load_state(os.path.join(args.output_dir, path))
+            
+            global_step = int(path.split("-")[1])
+
+            resume_global_step = global_step * args.gradient_accumulation_steps
+            first_epoch = global_step // num_update_steps_per_epoch
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
+            
+        
+            
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
+    
+    # caluate the elapsed time
+    elapsed_time = []
+    start = time.time()
+
+
+   
+    for epoch in range(first_epoch, args.num_train_epochs):
+        unet.train()
+        for step, (ref_image, image, masked_image, masks, audio_feature) in enumerate(train_data_loader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+            dataloader_time = time.time() - start
+            start = time.time()    
+            masks = masks.unsqueeze(1).unsqueeze(1).to(vae.device)
+            # """
+            # print("=============epoch:{0}=step:{1}=====".format(epoch,step))
+            # print("ref_image: ",ref_image.shape)
+            # print("masks: ", masks.shape)
+            # print("masked_image: ", masked_image.shape)
+            # print("audio feature: ", audio_feature.shape)
+            # print("image: ", image.shape)
+            # """
+            ref_image = preprocess_img_tensor(ref_image).to(vae.device)
+            image = preprocess_img_tensor(image).to(vae.device)
+            masked_image = preprocess_img_tensor(masked_image).to(vae.device)
+            
+            img_process_time = time.time() - start
+            start = time.time()
+            with accelerator.accumulate(unet):
+                vae = vae.half()
+                # Convert images to latent space 
+                latents = vae.encode(image.to(dtype=weight_dtype)).latent_dist.sample() # init image
+                latents = latents * vae.config.scaling_factor
+
+                # Convert masked images to latent space
+                masked_latents = vae.encode(
+                    masked_image.reshape(image.shape).to(dtype=weight_dtype)  # masked image
+                ).latent_dist.sample()
+                masked_latents = masked_latents * vae.config.scaling_factor
+                
+                # Convert ref images to latent space
+                ref_latents = vae.encode(
+                    ref_image.reshape(image.shape).to(dtype=weight_dtype)  # ref image
+                ).latent_dist.sample()
+                ref_latents = ref_latents * vae.config.scaling_factor
+                
+                vae_time = time.time() - start
+                start = time.time()
+
+                mask = torch.stack(
+                    [
+                        torch.nn.functional.interpolate(mask, size=(mask.shape[-1] // 8, mask.shape[-1] // 8))
+                        for mask in masks
+                    ]
+                )
+                mask = mask.reshape(-1, 1, mask.shape[-1], mask.shape[-1])
+
+
+                bsz = latents.shape[0]
+                # fix timestep for each image
+                timesteps = torch.tensor([0], device=latents.device)
+                # concatenate the latents with the mask and the masked latents
+                """
+                print("=============vae latents=====".format(epoch,step))
+                print("ref_latents: ",ref_latents.shape)
+                print("mask: ", mask.shape)
+                print("masked_latents: ", masked_latents.shape)
+                """
+
+                if unet_config['in_channels'] == 9:
+                    latent_model_input = torch.cat([mask, masked_latents, ref_latents], dim=1)
+                else:
+                    latent_model_input = torch.cat([masked_latents, ref_latents], dim=1)
+
+                audio_feature = audio_feature.to(dtype=weight_dtype)
+                audio_feature = pe(audio_feature)
+                
+                # Predict the noise residual
+                image_pred = unet(latent_model_input, 
+                                  timesteps, 
+                                  encoder_hidden_states=audio_feature).sample
+
+                if args.reconstruction: # decode the image from the predicted latents
+                    image_pred_img = (1 / vae_fp32.config.scaling_factor) * image_pred
+                    image_pred_img = vae_fp32.decode(image_pred_img).sample
+
+                    # Mask the top half of the image and calculate the loss only for the lower half of the image.
+                    image_pred_img = image_pred_img[:, :, image_pred_img.shape[2]//2:, :]
+                    image = image[:, :, image.shape[2]//2:, :]    
+                    loss_lip = F.l1_loss(image_pred_img.float(), image.float(), reduction="mean") # the loss of the decoded images
+                    loss_latents = F.l1_loss(image_pred.float(), latents.float(), reduction="mean") # the loss of the latents
+               
+                    loss = 2.0*loss_lip + loss_latents # add some weight to balance the loss
+                
+                else:
+                    loss = F.mse_loss(image_pred.float(), latents.float(), reduction="mean")
+#                   
+
+                unet_elapsed_time = time.time() - start
+                start = time.time()
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    params_to_clip = (
+                        itertools.chain(unet.parameters())
+                    )
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm) 
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+                
+                backward_elapsed_time = time.time() - start
+                start = time.time()
+                
+                if args.testing_speed is True and accelerator.is_main_process:
+                    elapsed_time.append(
+                        [dataloader_time, unet_elapsed_time, vae_time, backward_elapsed_time,img_process_time]
+                    )
+                
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+
+                if global_step % args.checkpointing_steps == 0:
+                    if accelerator.is_main_process:
+                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
+                        accelerator.save_state(save_path)
+                        logger.info(f"Saved state to {save_path}")
+                        
+
+                if global_step % args.validation_steps == 0:
+                    if accelerator.is_main_process:
+                        logger.info(
+                            f"Running validation... epoch={epoch}, global_step={global_step}"
+                        )
+                        print("===========start validation==========")
+                        # Use the helper function to check the data types for each model
+                        vae_new = vae.float()
+                        print_model_dtypes(accelerator.unwrap_model(vae_new), "VAE")
+                        print_model_dtypes(accelerator.unwrap_model(vae_fp32), "VAE_FP32")
+                        print_model_dtypes(accelerator.unwrap_model(unet), "UNET")
+
+                        print(f"weight_dtype: {weight_dtype}")
+                        print(f"epoch type: {type(epoch)}")
+                        print(f"global_step type: {type(global_step)}")
+                        validation(
+                            # vae=accelerator.unwrap_model(vae),
+                            vae=accelerator.unwrap_model(vae_new),
+                            vae_fp32=accelerator.unwrap_model(vae_fp32),
+                            unet=accelerator.unwrap_model(unet),
+                            unet_config=unet_config,
+                            # weight_dtype=weight_dtype,
+                            weight_dtype=torch.float32,
+                            epoch=epoch,
+                            global_step=global_step,
+                            val_data_loader=val_data_loader,
+                            output_dir = args.val_out_dir,
+                            whisper_model_type = args.whisper_model_type
+                        )
+                        logger.info(f"Saved samples to images/val")
+                    start = time.time()                    
+
+            logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0],
+                   "unet": unet_elapsed_time,
+                    "backward": backward_elapsed_time,
+                    "data": dataloader_time,
+                    "img_process":img_process_time,
+                    "vae":vae_time
+                   }
+            progress_bar.set_postfix(**logs)
+#             accelerator.log(logs, step=global_step)
+
+            accelerator.log(
+                {
+                    "loss/step_loss": logs["loss"],
+                    "parameter/lr": logs["lr"],
+                    "time/unet_forward_time": unet_elapsed_time,
+                    "time/unet_backward_time": backward_elapsed_time,
+                    "time/data_time": dataloader_time,
+                    "time/img_process_time":img_process_time,
+                    "time/vae_time": vae_time
+                },
+                step=global_step,
+            )
+
+            if global_step >= args.max_train_steps:
+                break
+
+        accelerator.wait_for_everyone()
+    accelerator.end_training()
+
+
+
+if __name__ == "__main__":
+    main()

train_codes/train.sh

@@ -0,0 +1,29 @@
+export VAE_MODEL="../models/sd-vae-ft-mse/"
+export DATASET="../data"
+export UNET_CONFIG="../models/musetalk/musetalk.json"
+
+accelerate launch train.py \
+--mixed_precision="fp16" \
+--unet_config_file=$UNET_CONFIG \
+--pretrained_model_name_or_path=$VAE_MODEL \
+--data_root=$DATASET \
+--train_batch_size=256 \
+--gradient_accumulation_steps=16 \
+--gradient_checkpointing \
+--max_train_steps=100000 \
+--learning_rate=5e-05 \
+--max_grad_norm=1 \
+--lr_warmup_steps=0 \
+--output_dir="output" \
+--val_out_dir='val' \
+--testing_speed \
+--checkpointing_steps=2000 \
+--validation_steps=2000 \
+--reconstruction \
+--resume_from_checkpoint="latest" \
+--use_audio_length_left=2 \
+--use_audio_length_right=2 \
+--whisper_model_type="tiny" \
+--train_json="../train.json" \
+--val_json="../val.json" \
+--lr_scheduler="cosine" \

train_codes/utils/model_utils.py

@@ -0,0 +1,129 @@
+import torch
+import torch.nn as nn
+
+import torch
+import torch.nn as nn
+import time
+import math
+from utils.utils import decode_latents, preprocess_img_tensor
+import os
+from PIL import Image
+from typing import Any, Dict, List, Optional, Tuple, Union
+from diffusers import (
+    AutoencoderKL,
+    UNet2DConditionModel,
+)
+from torch import Tensor, nn
+import logging
+import json
+
+RESIZED_IMG = 256
+
+class PositionalEncoding(nn.Module):
+    """
+    Transformer 中的位置编码（positional encoding）
+    """
+    def __init__(self, d_model=384, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        pe = torch.zeros(max_len, d_model)
+        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
+        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
+        pe[:, 0::2] = torch.sin(position * div_term)
+        pe[:, 1::2] = torch.cos(position * div_term)
+        pe = pe.unsqueeze(0)
+        self.register_buffer('pe', pe)
+
+    def forward(self, x):
+        b, seq_len, d_model = x.size()
+        pe = self.pe[:, :seq_len, :]
+        #print(b, seq_len, d_model)
+        x = x + pe.to(x.device)
+        return x
+
+def validation(vae: torch.nn.Module,
+               vae_fp32: torch.nn.Module,
+      unet:torch.nn.Module, 
+      unet_config,
+      weight_dtype: torch.dtype,
+      epoch: int,
+      global_step: int,
+      val_data_loader,
+      output_dir,
+      whisper_model_type,
+      UNet2DConditionModel=UNet2DConditionModel
+     ):
+    
+     # Get the validation pipeline
+    unet_copy = UNet2DConditionModel(**unet_config)
+    
+    unet_copy.load_state_dict(unet.state_dict())
+    unet_copy.to(vae.device).to(dtype=weight_dtype)
+    unet_copy.eval()
+    
+    if whisper_model_type == "tiny":
+        pe = PositionalEncoding(d_model=384)
+    elif whisper_model_type == "largeV2":
+        pe = PositionalEncoding(d_model=1280)
+    elif whisper_model_type == "tiny-conv":
+        pe = PositionalEncoding(d_model=384)
+        print(f" whisper_model_type: {whisper_model_type} Validation does not need PE")
+    else:
+        print(f"not support whisper_model_type {whisper_model_type}")
+    pe.to(vae.device, dtype=weight_dtype)
+    
+    start = time.time()
+    with torch.no_grad():
+        for step, (ref_image, image, masked_image, masks, audio_feature) in enumerate(val_data_loader):
+
+
+            masks = masks.unsqueeze(1).unsqueeze(1).to(vae.device)
+            ref_image = preprocess_img_tensor(ref_image).to(vae.device)
+            image = preprocess_img_tensor(image).to(vae.device)
+            masked_image = preprocess_img_tensor(masked_image).to(vae.device)
+
+             # Convert images to latent space 
+            latents = vae.encode(image.to(dtype=weight_dtype)).latent_dist.sample() # init image
+            latents = latents * vae.config.scaling_factor
+            # Convert masked images to latent space
+            masked_latents = vae.encode(
+                masked_image.reshape(image.shape).to(dtype=weight_dtype)  # masked image
+            ).latent_dist.sample()
+            masked_latents = masked_latents * vae.config.scaling_factor
+            # Convert ref images to latent space
+            ref_latents = vae.encode(
+                ref_image.reshape(image.shape).to(dtype=weight_dtype)  # ref image
+            ).latent_dist.sample()
+            ref_latents = ref_latents * vae.config.scaling_factor
+
+            mask = torch.stack(
+                [
+                    torch.nn.functional.interpolate(mask, size=(mask.shape[-1] // 8, mask.shape[-1] // 8))
+                    for mask in masks
+                ]
+            )
+            mask = mask.reshape(-1, 1, mask.shape[-1], mask.shape[-1])
+            bsz = latents.shape[0]
+            timesteps = torch.tensor([0], device=latents.device)
+
+            if unet_config['in_channels'] == 9:
+                latent_model_input = torch.cat([mask, masked_latents, ref_latents], dim=1)
+            else:
+                latent_model_input = torch.cat([masked_latents, ref_latents], dim=1)
+
+            image_pred = unet_copy(latent_model_input, timesteps, encoder_hidden_states = audio_feature).sample
+
+            image = Image.new('RGB', (RESIZED_IMG*4, RESIZED_IMG))
+            image.paste(decode_latents(vae_fp32,masked_latents), (0, 0))
+            image.paste(decode_latents(vae_fp32, ref_latents), (RESIZED_IMG, 0))
+            image.paste(decode_latents(vae_fp32, latents), (RESIZED_IMG*2, 0))
+            image.paste(decode_latents(vae_fp32, image_pred), (RESIZED_IMG*3, 0))
+
+            val_img_dir = f"images/{output_dir}/{global_step}"
+            if not os.path.exists(val_img_dir):
+                os.makedirs(val_img_dir)
+            image.save('{0}/val_epoch_{1}_{2}_image.png'.format(val_img_dir, global_step,step))
+
+            print("valtion in step:{0}, time:{1}".format(step,time.time()-start))
+
+        print("valtion_done in epoch:{0}, time:{1}".format(epoch,time.time()-start))
+    

train_codes/utils/utils.py

@@ -0,0 +1,74 @@
+import matplotlib.pyplot as plt
+import PIL
+from PIL import Image
+import numpy as np
+
+import torch
+import torch.nn.functional as F
+import torchvision.transforms.functional as TF
+from einops import rearrange
+
+import torch
+import torchvision.transforms as transforms
+
+from diffusers import AutoencoderKL
+import matplotlib.pyplot as plt
+import PIL
+import os
+import cv2
+from glob import glob
+
+
+def preprocess_img_tensor(image_tensor):
+    # 假设输入是一个形状为 (N, C, H, W) 的 PyTorch 张量
+    N, C, H, W = image_tensor.shape
+    # 计算新的宽度和高度，使其为 32 的整数倍
+    new_w = W - W % 32
+    new_h = H - H % 32
+    # 使用 torchvision.transforms 库中的方法进行缩放和重采样
+    transform = transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+    # 对每个图像应用变换，并将结果存储在一个新的张量中
+    preprocessed_images = torch.empty((N, C, new_h, new_w), dtype=torch.float32)
+    for i in range(N):
+        # 使用 F.interpolate 替换 transforms.Resize
+        resized_image = F.interpolate(image_tensor[i].unsqueeze(0), size=(new_h, new_w), mode='bilinear', align_corners=False)
+        preprocessed_images[i] = transform(resized_image.squeeze(0))
+
+    return preprocessed_images
+
+
+def prepare_mask_and_masked_image(image_tensor, mask_tensor):
+    # 假设输入 image_tensor 的形状为 [C, H, W]，输入 mask_tensor 的形状为 [H, W]
+#     # 对图像张量进行归一化
+    image_tensor_ori = (image_tensor.to(dtype=torch.float32) / 127.5) - 1.0
+#     # 对遮罩张量进行归一化和二值化
+#     mask_tensor = (mask_tensor.to(dtype=torch.float32) / 255.0).unsqueeze(0)
+    mask_tensor[mask_tensor < 0.5] = 0
+    mask_tensor[mask_tensor >= 0.5] = 1
+    # 创建遮罩后的图像
+    masked_image_tensor = image_tensor * (mask_tensor > 0.5)
+
+    return mask_tensor, masked_image_tensor
+
+
+def encode_latents(vae, image):
+#     init_image = preprocess_image(image) 
+    init_latent_dist = vae.encode(image.to(vae.dtype)).latent_dist
+    init_latents = 0.18215 * init_latent_dist.sample()
+    return init_latents
+
+def decode_latents(vae, latents, ref_images=None):
+    latents = (1/  0.18215) * latents
+    image = vae.decode(latents.to(vae.dtype)).sample
+    image = (image / 2 + 0.5).clamp(0, 1)
+    image = image.detach().cpu().permute(0, 2, 3, 1).float().numpy()
+    image = (image * 255).round().astype("uint8")
+    if ref_images is not None:
+        ref_images = ref_images.detach().cpu().permute(0, 2, 3, 1).float().numpy()
+        ref_images = (ref_images * 255).round().astype("uint8")
+        h = image.shape[1]
+        image[:, :h//2] = ref_images[:, :h//2]
+    image = [Image.fromarray(im) for im in image]
+
+    return image[0]
+