DevanshL/training-free-amodal-grasping-data

Experiment data: Box or Round? Training-Free Primitive Completion of Occluded Objects

Every result file behind the paper and the code at github.com/DevanshL/training-free-amodal-grasping: LIBERO-Occ scenes with ground truth, SAM-2 masks, VLM replies, fits of the method and every baseline, robustness runs, Amodal3R results and robot pick episodes, for the 202-scene benchmark, the 58-scene development split and the 177-scene held-out split; and the real RGB-D evaluation on YCB-Video (BOP format): scenes built from its test images, masks, VLM replies, fits, baselines and Amodal3R results for 4 development and 8 test scenes.

Use

In a clone of the GitHub repository (after README "After cloning", steps 1–4):

bash setup/download_data.sh     # downloads these archives, checks SHA256SUMS, unpacks into data/
bash setup/verify_results.sh    # regenerates all tables and figures and checks them against the published ones

Files

One tar archive per data/ folder (unpack with tar -xf <name>.tar -C data). SHA256SUMS holds the checksums.

Archive	Contents
occ_benchmark.tar	202-scene benchmark: scenes, masks, VLM replies, fits, baselines, robustness, Amodal3R, end-to-end
occ_dev.tar	58-scene development split (every later design choice was made here)
occ_fresh.tar	177-scene held-out split, evaluated once with the final method
pick_test.tar	Robot pick episodes on the benchmark (5 conditions per scene)
graspability*.tar	Graspability study (and its two superseded variants, which the report compares)
controller_heldout*.tar	Scripted-controller tuning and test splits
ycbv.tar	YCB-Video ground-truth check (gt_check.md, gt_check.json) and the seeded development/test scene split
occ_ycbv_dev.tar	YCB-Video development scenes (4 scenes, 120 objects): every real-data choice was made here
occ_ycbv_test.tar	YCB-Video test scenes (8 scenes, 264 objects), evaluated once

The YCB-Video images themselves come from the BOP benchmark (bop-benchmark/ycbv on HuggingFace); the scene files in occ_ycbv_*.tar hold the RGB-D images and ground truth derived from them (see pipeline/REPRODUCE.md, section 8).

Licence

The LIBERO-Occ scenes are rendered from LIBERO and LIBERO-Occ, and the YCB-Video scenes are derived from YCB-Video (Xiang et al., RSS 2018) as distributed by the BOP benchmark; all three are under the MIT licence. Their notices follow.

MIT License — Copyright (c) 2023 Lifelong Robot Learning (LIBERO)

MIT License — Copyright (c) 2026 LIBERO-Occ authors (LIBERO-Occ)

MIT License — YCB-Video dataset (Y. Xiang, T. Schmidt, V. Narayanan, D. Fox), BOP version (bop.felk.cvut.cz)

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED.