Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

update csvutil for vidprom

Zangwei Zheng24c707bfb941
diff --git a/tools/datasets/README.md b/tools/datasets/README.md
index 0118c8f..ffa9590 100644
--- a/tools/datasets/README.md
+++ b/tools/datasets/README.md
@@ -13,6 +13,12 @@ The training data should be provided in a CSV file with the following format:
This dataset comprises 130M text-video pairs. You can download the dataset and prepare it for training according to [the dataset repository's instructions](https://github.com/daooshee/HD-VG-130M). There is a README.md file in the Google Drive link that provides instructions on how to download and cut the videos. For this version, we directly use the dataset provided by the authors.
+## VidProM
+
+```bash
+python -m tools.datasets.convert_dataset vidprom VIDPROM_FOLDER --info VidProM_semantic_unique.csv
+```
+
## Demo Dataset
You can use ImageNet and UCF101 for a quick demo. After downloading the datasets, you can use the following command to prepare the csv file for the dataset:
diff --git a/tools/datasets/convert_dataset.py b/tools/datasets/convert_dataset.py
index 4ff904f..4fa8719 100644
--- a/tools/datasets/convert_dataset.py
+++ b/tools/datasets/convert_dataset.py
@@ -51,16 +51,38 @@ def process_ucf101(root, split):
print(f"Saved {len(samples)} samples to {output}.")
+def process_vidprom(root, info):
+ root = os.path.expanduser(root)
+ video_lists = get_filelist(root)
+ video_set = set(video_lists)
+ # read info csv
+ with open(info, "r") as f:
+ reader = csv.reader(f)
+ lines = list(reader)
+ infos = [(os.path.join(root, f"pika-{x[0]}.mp4"), x[1]) for x in lines[1:]]
+ samples = [x for x in infos if x[0] in video_set]
+ output = "vidprom.csv"
+
+ with open(output, "w") as f:
+ writer = csv.writer(f)
+ writer.writerows(samples)
+
+ print(f"Saved {len(samples)} samples to {output}.")
+
+
if __name__ == "__main__":
parser = argparse.ArgumentParser()
- parser.add_argument("dataset", type=str, choices=["imagenet", "ucf101"])
+ parser.add_argument("dataset", type=str, choices=["imagenet", "ucf101", "vidprom"])
parser.add_argument("root", type=str)
parser.add_argument("--split", type=str, default="train")
+ parser.add_argument("--info", type=str, default=None)
args = parser.parse_args()
if args.dataset == "imagenet":
process_imagenet(args.root, args.split)
elif args.dataset == "ucf101":
process_ucf101(args.root, args.split)
+ elif args.dataset == "vidprom":
+ process_vidprom(args.root, args.info)
else:
raise ValueError("Invalid dataset")
diff --git a/tools/datasets/csvutil.py b/tools/datasets/csvutil.py
index 4bbd22d..990f665 100644
--- a/tools/datasets/csvutil.py
+++ b/tools/datasets/csvutil.py
@@ -24,6 +24,21 @@ def get_video_length(path):
return int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+def shard_csv(data, shard, input_path):
+ n = len(data)
+ num_shard = n // shard
+ for i in range(shard):
+ start = i * num_shard
+ end = (i + 1) * num_shard if i != shard - 1 else n
+ file_name, ext = os.path.splitext(input_path)
+ output_path = os.path.join(os.path.join(os.path.dirname(file_name)), f"{os.path.basename(file_name)}_{i}{ext}")
+ with open(output_path, "w") as f:
+ writer = csv.writer(f)
+ writer.writerows(data[start:end])
+ print(f"Saved {end - start} samples to {output_path}.")
+ print("Done.")
+2
+
def main(args):
input_path = args.input
output_path = args.output
@@ -49,11 +64,16 @@ def main(args):
data = list(reader)
print("Number of videos before filtering:", len(data))
+ if args.shard is not None:
+ shard_csv(data, args.shard, input_path)
+ return
+
data_new = []
for i, row in tqdm(enumerate(data)):
path = row[0]
caption = row[1]
- n_frames = int(row[2])
+ if len(row) >= 3:
+ n_frames = int(row[2])
if args.fmin is not None and n_frames < args.fmin:
continue
if args.fmax is not None and n_frames > args.fmax:
@@ -72,7 +92,10 @@ def main(args):
row[0] = os.path.join(args.root, path)
if args.relength:
n_frames = get_video_length(row[0])
- row[2] = n_frames
+ if len(row) < 3:
+ row.append(n_frames)
+ else:
+ row[2] = n_frames
data_new.append(row)
print("Number of videos after filtering:", len(data_new))
@@ -92,5 +115,6 @@ if __name__ == "__main__":
parser.add_argument("--remove-empty-caption", action="store_true")
parser.add_argument("--remove-caption-prefix", action="store_true")
parser.add_argument("--relength", action="store_true")
+ parser.add_argument("--shard", type=int, default=None)
args = parser.parse_args()
main(args)