File size: 6,804 Bytes
33569f9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 | # preprocess_dataset.py
import argparse
import json
import multiprocessing as mp
import os
import torch
from datasets import Dataset, DatasetDict
from tqdm import tqdm
from transformers import AutoProcessor
from vision_process import process_vision_info
def parse_args():
parser = argparse.ArgumentParser(
description="Preprocess video dataset for Qwen-VL model"
)
parser.add_argument(
"--model_name",
type=str,
default="/share/pretrain/mllm/Qwen2.5-VL-7B-Instruct",
help="Path to the pretrained model",
)
parser.add_argument(
"--dataset",
type=str,
default="charades",
help="Dataset name to be preprocessed",
)
parser.add_argument(
"--train_data_path",
type=str,
default="./Charades/charades_annotation/train.json",
help="Path to the training data JSON file",
)
parser.add_argument(
"--eval_data_path",
type=str,
default="./Charades/charades_annotation/val.json",
help="Path to the evaluation data JSON file",
)
parser.add_argument(
"--video_folder",
type=str,
default="./Charades/Charades_v1",
help="Path to the folder containing video files",
)
parser.add_argument(
"--output_dir",
type=str,
default=None,
help="Output directory path. If None, it will be created based on dataset and max_pix values",
)
parser.add_argument(
"--max_pix_size", type=int, default=3584, help="Maximum pixel size"
)
parser.add_argument(
"--min_pix_size", type=int, default=16, help="Minimum pixel size"
)
parser.add_argument(
"--num_workers",
type=int,
default=16,
help="Number of worker processes for multiprocessing",
)
return parser.parse_args()
def preprocess_single_video(task_args): # Accept task arguments as a single tuple/list
video_path, processor, max_pixels, min_pixels, example_output_dir = (
task_args # Unpack task args
)
try:
if os.path.exists(example_output_dir):
return {"preprocessed_path": example_output_dir, "status": "success"}
else:
image_inputs, video_inputs, video_kwargs, fps_inputs = (
preprocess_video_inner(video_path, processor, max_pixels, min_pixels)
)
os.makedirs(example_output_dir, exist_ok=True)
torch.save(
video_inputs, os.path.join(example_output_dir, "video_inputs.pt")
)
with open(os.path.join(example_output_dir, "video_kwargs.json"), "w") as f:
json.dump(video_kwargs, f)
return {
"preprocessed_path": example_output_dir,
"status": "success",
}
except Exception as e:
print(
f"Warning: Preprocessing failed for video {video_path}, skipping. Error: {e}"
)
return {"video_path": video_path, "status": "failed", "error": str(e)}
def preprocess_video_inner(video_path, processor, max_pixels, min_pixels):
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": video_path,
"total_pixels": max_pixels,
"min_pixels": min_pixels,
},
],
},
]
image_inputs, video_inputs, video_kwargs = process_vision_info(
[messages], return_video_kwargs=True
)
fps_inputs = video_kwargs["fps"]
return image_inputs, video_inputs, video_kwargs, fps_inputs
def process_split(
file_path,
split_name,
video_folder,
output_dir,
max_pixels,
min_pixels,
processor,
num_workers=8,
):
with open(file_path, "r") as f:
data = json.load(f)
tasks = []
for video_id, video_data in data.items():
video_filename_base = video_id
video_path = None
for ext in ["mp4", "mkv", "webm"]:
candidate_path = os.path.join(video_folder, f"{video_filename_base}.{ext}")
if os.path.isfile(candidate_path):
video_path = candidate_path
break
if video_path is None:
print(f"Warning: Video file not found for ID: {video_id}")
continue
example_output_dir = os.path.join(output_dir, video_id)
tasks.append(
(video_path, processor, max_pixels, min_pixels, example_output_dir)
) # Prepare task arguments as tuples
pbar = tqdm(
total=len(tasks), desc=f"Preprocessing {split_name} split"
) # Initialize progress bar in main process
with mp.Pool(processes=num_workers) as pool:
results = pool.imap_unordered(
preprocess_single_video, tasks
) # Use imap_unordered for unordered results, potentially faster
successful_examples = []
failed_count = 0
for result in results: # Iterate through results to update progress bar
pbar.update(1)
if result["status"] == "success":
successful_examples.append(result)
else:
failed_count += 1
# Optionally log failed videos and errors
pbar.close() # Close progress bar after processing
print(
f"Preprocessing for split '{split_name}' finished. Failed videos: {failed_count}, Successful videos: {len(successful_examples)}"
)
return Dataset.from_list(successful_examples)
def preprocess_dataset_and_save(
train_data_path, video_folder, output_dir, max_pixels, min_pixels, num_workers=8
):
processor = AutoProcessor.from_pretrained(MODEL_NAME)
os.makedirs(output_dir, exist_ok=True)
train_dataset = process_split(
train_data_path,
"train",
video_folder,
output_dir,
max_pixels,
min_pixels,
processor,
num_workers,
)
return DatasetDict({"train": train_dataset})
if __name__ == "__main__":
args = parse_args()
MODEL_NAME = args.model_name
# Calculate pixel values
max_pixels = args.max_pix_size * 28 * 28
min_pixels = args.min_pix_size * 28 * 28
# Setup output directory
if args.output_dir is None:
output_dir = f"./{args.dataset}_preprocessed_data_maxpix_{args.max_pix_size}"
else:
output_dir = args.output_dir
print("output_dir", output_dir)
dataset_dict = preprocess_dataset_and_save(
args.train_data_path,
args.video_folder,
output_dir,
max_pixels,
min_pixels,
num_workers=args.num_workers,
)
print("Preprocessing complete. Datasets saved to:", output_dir)
print(dataset_dict)
|