athenacon/

3DIEBench-T

This dataset was created and released alongside the paper EquiCaps: Predictor-Free Pose-Aware Pre-Trained Capsule Networks.

3DIEBench-T (3D Invariant Equivariant Benchmark–Translated) is a synthetic vision benchmark designed to evaluate invariant and equivariant representations under 3D transformations. Extending the original 3DIEBench rotation benchmark, 3DIEBench-T incorporates translations to the existing rotations by… See the full description on the dataset page: https://huggingface.co/datasets/athenacon/3DIEBench-T.