RoboKeyBench/

RoboKeyBench_v2

merged.json & vqa_dataset.json:

VQA (Visual Question Answering) question-answer pairs for Multimodal Large Language Models (MLLMs)

render_x90_y0_z0.png:

Rendered images aligned with VQA tasks Filename convention: render_x[θ]_y[φ]_z[ψ], where: θ = 90° rotation about x-axis φ = 0° rotation about y-axis ψ = 0° rotation about z-axis (Euler angles in ZYX convention)

keypoints.json & 2D_keypoints.json:

3D/2D keypoint coordinate sets used for image rendering Coordinate system: 3D… See the full description on the dataset page: https://huggingface.co/datasets/RoboKeyBench/RoboKeyBench_v2.