Skip to content

Dataset · bop-benchmark

HOT3D clips collection for 3D object and hand pose estimation

bop-benchmark/hot3d

Curated video fragments and fisheye camera images from the HOT3D benchmark for 3D vision tasks.

What's inside

The repository contains HOT3D-Clips, a set of selected sub-sequences from the HOT3D dataset (from 100k to 1 million elements), including images and text metadata. To work with the data, including fisheye camera distortion removal and rendering, it is recommended to use the specialized HOT3D Toolkit. The dataset is associated with the BOP 2024 benchmark report.

What you can do with it

  • Training and evaluation of 3D object 6D pose estimation models
  • Tracking hand-object interactions in 3D environments
  • Calibration and processing of images from wide-angle fisheye cameras
Size
100K–1M rows
Downloads this month
102.7K
Likes
22

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("bop-benchmark/hot3d")