File size: 5,853 Bytes
bbaf249 |
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 |
from time_r1.utils.qwen_vl_utils_prepare import floor_by_factor, FRAME_FACTOR, smart_resize, ceil_by_factor
import decord
import torch
import os
import tqdm
import glob
import multiprocessing
from torchvision import io, transforms
from torchvision.transforms import InterpolationMode
from functools import partial
import json
os.environ['DECORD_EOF_RETRY_MAX'] = '20480'
# ================= 配置区域 =================
DATA_path = "/data/shuimu.chen/Video-R1-data/Video-R1-103K_sft_candidates_60k_id.json"
# 定义源目录根路径 (原始视频所在的根目录)
SOURCE_ROOT = "/data/shuimu.chen/Video-R1-data"
# 定义目标目录根路径 (你想保存缓存的新目录)
TARGET_ROOT = "/data/shuimu.chen/Video-R1-data_cache_60_sft"
FPS_MIN_FRAMES = 4
FPS_MAX_FRAMES = 60
FRAME_FACTOR = 2
# ===========================================
def get_video_tensor(video_path, target_fps=2, image_factor=28, min_pixels=28 * 28 * 4, max_pixels=28 * 28 * 128):
"""
将视频以固定帧率提前抽帧、解码保存为tensor
"""
vr = decord.VideoReader(video_path)
total_frames, video_fps = len(vr), vr.get_avg_fps()
min_nframes = ceil_by_factor(FPS_MIN_FRAMES, FRAME_FACTOR)
max_frames = floor_by_factor(min(FPS_MAX_FRAMES, total_frames), FRAME_FACTOR)
nframes = total_frames / video_fps * target_fps
nframes = min(min(max(nframes, min_nframes), max_frames), total_frames)
nframes = floor_by_factor(nframes, FRAME_FACTOR)
frame_idx = torch.linspace(0, total_frames - 1, nframes).round().long().tolist()
frame_tensor = vr.get_batch(frame_idx).asnumpy()
frame_tensor = torch.tensor(frame_tensor).permute(0, 3, 1, 2)
sample_fps = nframes / max(total_frames, 1e-6) * video_fps
height, width = frame_tensor.shape[2], frame_tensor.shape[3]
resized_height, resized_width = smart_resize(
height,
width,
factor=image_factor,
min_pixels=min_pixels,
max_pixels=max_pixels,
)
frame_tensor = transforms.functional.resize(
frame_tensor,
[resized_height, resized_width],
interpolation=InterpolationMode.BICUBIC,
antialias=True,
)
frame_cache = {
"frame_tensor": frame_tensor,
"fps": sample_fps,
}
return frame_cache
def process_single_video(video_path, target_fps=1, image_factor=28, min_pixels=28*28*128, max_pixels=28*28*256):
# 1. 计算新的保存路径
try:
# 计算相对路径:例如 NeXT-QA/NextQA/NExTVideo/1012/8702139498.mp4
rel_path = os.path.relpath(video_path, SOURCE_ROOT)
# 拼接目标路径:/data/.../Video-R1-data_cache_60/NeXT-QA/.../8702139498.mp4.frame_cache
save_path = os.path.join(TARGET_ROOT, rel_path + ".frame_cache")
except ValueError:
# 如果 video_path 不在 SOURCE_ROOT 下(比如路径写错了),回退到保持原文件名保存在 TARGET_ROOT 根目录下,或者报错
# 这里选择简单报错返回
return f"Error Path: {video_path} is not under {SOURCE_ROOT}"
# 2. 检查是否已处理 (断点续传)
if os.path.exists(save_path):
# print(f"Skipping {save_path} (Already exists)")
return None
# print(f"Processing {video_path} -> {save_path}...")
try:
# 3. 确保目标目录存在
os.makedirs(os.path.dirname(save_path), exist_ok=True)
# 4. 处理视频
frame_cache = get_video_tensor(video_path, target_fps, image_factor, min_pixels, max_pixels)
# 5. 保存
torch.save(frame_cache, save_path)
# print(f"Saved: {save_path}")
return None
except Exception as e:
return f"Error: {video_path} -> {str(e)}"
def prepare_frame_cache(video_root=None, dataset_path=None, num_workers=8, target_fps=2, overwrite=False, image_factor=28, min_pixels=28*28*128, max_pixels=28*28*256):
dataset_path = DATA_path
print(f"Source Root: {SOURCE_ROOT}")
print(f"Target Root: {TARGET_ROOT}")
if dataset_path is not None:
with open(dataset_path, 'r', encoding='utf-8') as f:
video_data = json.load(f)
video_paths = set()
for v in video_data:
# 兼容不同的 key
path = v.get("video_path") or v.get("video") or v.get("path")
if path:
# 确保是绝对路径
if not os.path.isabs(path) and video_root:
path = os.path.join(video_root, path)
video_paths.add(path)
video_list = list(video_paths)
else:
# 如果没传 json,就扫描目录(通常不推荐,因为慢)
video_list = glob.glob(os.path.join(SOURCE_ROOT, "**", "*.mp4"), recursive=True)
print(f"Starting processing {len(video_list)} videos...")
failed_videos = []
# 使用 partial 传递参数
func = partial(process_single_video, target_fps=target_fps, image_factor=image_factor, min_pixels=min_pixels, max_pixels=max_pixels)
with multiprocessing.Pool(processes=num_workers) as pool:
# 使用 imap_unordered + tqdm 显示进度
for result in tqdm.tqdm(pool.imap_unordered(func, video_list), total=len(video_list)):
if result is not None:
failed_videos.append(result)
print("\n" + "="*30)
print("Processing Complete.")
if failed_videos:
print(f"❌ {len(failed_videos)} videos failed to process. Check 'failed_log.txt'")
# 将失败的记录保存下来,方便排查
with open("failed_log.txt", "w") as f:
for err in failed_videos:
f.write(err + "\n")
print(err)
else:
print("✅ All videos processed successfully.")
if __name__ == "__main__":
import fire
fire.Fire(prepare_frame_cache) |