README.md

metadata

license: cc-by-4.0
task_categories:
  - text-to-video
language:
  - en
tags:
  - text-to-video
  - Video Generative Model Training
  - Text-to-Video Diffusion Model Training
  - prompts
pretty_name: OpenVid-1M
size_categories:
  - 10K<n<100K
dataset_info:
  features:
    - name: video
      dtype: video
    - name: caption
      dtype: string
    - name: aesthetic_score
      dtype: float32
    - name: motion_score
      dtype: float32
    - name: temporal_consistency_score
      dtype: float32
    - name: camera_motion
      dtype: string
    - name: frame
      dtype: int32
    - name: fps
      dtype: float32
    - name: seconds
      dtype: float32
    - name: part_id
      dtype: int32

Combination of part_id's from bigdata-pw/OpenVid-1M and video data from nkp37/OpenVid-1M.

This is a 10k video split of the original dataset for faster iteration during testing. The split was obtained by filtering on aesthetic and motion scores by iteratively increasing their values until there were at most 1000 videos. Only videos containing between 80 and 240 frames were considered.

from datasets import load_dataset, disable_caching, DownloadMode
from torchcodec.decoders import VideoDecoder

# disable_caching()

def decode_float(sample):
    return float(sample.decode("utf-8"))

def decode_int(sample):
    return int(sample.decode("utf-8"))

def decode_str(sample):
    return sample.decode("utf-8")

def decode_video(sample):
    decoder = VideoDecoder(sample)
    return decoder[:1024]

def decode_batch(batch):
    decoded_sample = {
        "__key__": batch["__key__"],
        "__url__": batch["__url__"],
        "video": list(map(decode_video, batch["video"])),
        "caption": list(map(decode_str, batch["caption"])),
        "aesthetic_score": list(map(decode_float, batch["aesthetic_score"])),
        "motion_score": list(map(decode_float, batch["motion_score"])),
        "temporal_consistency_score": list(map(decode_float, batch["temporal_consistency_score"])),
        "camera_motion": list(map(decode_str, batch["camera_motion"])),
        "frame": list(map(decode_int, batch["frame"])),
        "fps": list(map(decode_float, batch["fps"])),
        "seconds": list(map(decode_float, batch["seconds"])),
        "part_id": list(map(decode_int, batch["part_id"])),
    }
    return decoded_sample

ds = load_dataset("finetrainers/OpenVid-10k-split", split="train", download_mode=DownloadMode.REUSE_DATASET_IF_EXISTS)
ds.set_transform(decode_batch)
iterator = iter(ds)

for i in range(10):
    data = next(iterator)
    breakpoint()

Environment tested:

- huggingface_hub version: 0.25.2
- Platform: macOS-15.3.1-arm64-arm-64bit
- Python version: 3.11.10
- Running in iPython ?: No
- Running in notebook ?: No
- Running in Google Colab ?: No
- Running in Google Colab Enterprise ?: No
- Token path ?: /Users/aryanvs/Desktop/huggingface/token
- Has saved token ?: True
- Who am I ?: a-r-r-o-w
- Configured git credential helpers: osxkeychain
- FastAI: N/A
- Tensorflow: N/A
- Torch: 2.6.0
- Jinja2: 3.1.4
- Graphviz: N/A
- keras: N/A
- Pydot: N/A
- Pillow: 10.4.0
- hf_transfer: 0.1.8
- gradio: 5.6.0
- tensorboard: N/A
- numpy: 1.26.4
- pydantic: 2.10.1
- aiohttp: 3.10.10
- ENDPOINT: https://huggingface.co
- HF_HUB_CACHE: /Users/aryanvs/Desktop/huggingface/hub
- HF_ASSETS_CACHE: /Users/aryanvs/Desktop/huggingface/assets
- HF_TOKEN_PATH: /Users/aryanvs/Desktop/huggingface/token
- HF_HUB_OFFLINE: False
- HF_HUB_DISABLE_TELEMETRY: False
- HF_HUB_DISABLE_PROGRESS_BARS: None
- HF_HUB_DISABLE_SYMLINKS_WARNING: False
- HF_HUB_DISABLE_EXPERIMENTAL_WARNING: False
- HF_HUB_DISABLE_IMPLICIT_TOKEN: False
- HF_HUB_ENABLE_HF_TRANSFER: True
- HF_HUB_ETAG_TIMEOUT: 10
- HF_HUB_DOWNLOAD_TIMEOUT: 10