datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
kibitzer-chessbot-dense-12mStepCount-Dense
StepCount-Dense
StepCount-Bench v3 的 eval-ready 子集。该仓库用于研究与非商业评测，下载采用 manual gated approval。
数据概览
样本数：1229
数量范围：11–50
每个样本包含：final count、归一化 gt_points、可选 gt_bbox、domain、scene level、ID/OOD category split、source provenance。
数据源：crowdhuman=300, dense200=67, fsc147_test=1, fscd_147=300, fscd_lvis=300, sku110k=9, visdrone_det=252
领域：aerial=252, cluttered_multiclass=300, crowd=300, everyday=301, generic_dense=67, retail=9
坐标契约：gt_points=[x,y]、gt_bbox=[x1,y1,x2… See the full description on the dataset page: https://huggingface.co/datasets/SI-Lab/StepCount-Dense.densesteer-repro-bundleDenseSpeech
DenseSpeech Dataset
Dense Speech Video Grounding Dataset for video grounding tasks with dense speech annotations.
Dataset Structure
Dense_Speech_Video_Grounding/
├── metadata/
│ ├── train.json # Training set annotations
│ ├── val.json # Validation set annotations
│ ├── test.json # Test set annotations
│ └── merged.json # Merged annotations
├── features/
│ └── {video_id}/ # Feature files for each video
│ ├── rgb.npy
│ ├──… See the full description on the dataset page: https://huggingface.co/datasets/MarvinP/DenseSpeech.