datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
FIRM-Video
FIRM-Video-SFT-90K
This repository releases the 90K SFT data for FIRM-Video.
The dataset covers three key evaluation dimensions:
Instruction Following (IF): whether the generated video accurately follows the text prompt.
Visual Quality (VQ): perceptual and technical quality, including clarity, sharpness, artifacts, flicker, and overall visual fidelity.
World Coherence (WC): whether the video is coherent with commonsense, temporal consistency, physical plausibility, and… See the full description on the dataset page: https://huggingface.co/datasets/VisionXLab/FIRM-Video.first_framesfire-laion-real-5k