Research map / 3D vision / Human pose estimation
TARGO and TARGO-Net: Benchmarking Target-Driven Object Grasping Under Occlusions
Yan Xia, Ran Ding, Ziyuan Qin, Guanqi Zhan, Kaichen Zhou, Long Yang, Hao Dong, Daniel Cremers
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add shape completion module of the target (ablation) | TARGO-Net · TARGO-Synthetic, occlusion above 0.4 | grasp success rate: up to 18% decrease without it | Improved |
| use completed target and completed target scene instead of partial target and raw scene (ablation) | TARGO-Net · TARGO-Synthetic, high occlusion | grasp success rate: approximately 20% drop for PT with RS after [0.4, 0.5) | Improved |
| use completed target (CT) with completed scene instead of partial target with CTS (ablation) | TARGO-Net · TARGO-Synthetic, high occlusion | grasp success rate: about 10% decline with partial target; about 12% decline with raw scene | Improved |
| add single-scene data augmentation to training (ablation) | VGN, GIGA-Aff and TARGO-Net · TARGO-Synthetic test | grasp success rate | Improved |
| use shape completion instead of scene completion | TARGO-Net vs GIGA · TARGO-Synthetic target completion | Chamfer L1 (x1000): 8.722 -> 2.443 | Improved |
| use shape completion instead of scene completion | TARGO-Net vs GIGA · TARGO-Synthetic target completion | IoU (%): 78.77 -> 93.99 | Improved |
| use cross-attention Transformer backbone instead of 3D CNN with concat or add fusion (ablation) | TARGO-Net w/o SC · TARGO-Synthetic, Gaussian noise sigma 0.005 | grasp success rate: 2% better than second best, 5% better than worst | Improved |
| use two cross-attention Transformer layers instead of one or three (ablation) | TARGO-Net · TARGO-Synthetic test | grasp success rate | Improved |
| switch from simulation to real-world test | TARGO-Net · Real-world vs simulation, high occlusion | grasp success rate drop: 7% -> 14% | Hurt |
| replace GIGA with TARGO-Net | real-world grasping · Real-world Hard occlusion [0.6,0.9) | grasp success rate (%): 30.0 -> 66.7 | Improved |
| increase number of occluders from 2 to 6 | GIGA · TARGO-Synthetic, occlusion [0.7, 0.8) | grasp success rate: decreases by 15% | Hurt |
| increase number of occluders from 2 to 6 | TARGO-Net · TARGO-Synthetic, occlusion [0.2, 0.3) | grasp success rate: increases by 14% | Improved |
Closest papers by meaning
- ZeroGrasp: Zero-Shot Shape Reconstruction Enabled Robotic Grasping (CVPR 2025 · similarity 0.88)
- Generalizing 6-DoF Grasp Detection via Domain Prior Knowledge (CVPR 2024 · similarity 0.85)
- CADGrasp: Learning Contact and Collision Aware General Dexterous Grasping in Cluttered Scenes (NeurIPS 2025 · similarity 0.85)
- Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp Generation (NeurIPS 2025 · similarity 0.84)
- UniGraspTransformer: Simplified Policy Distillation for Scalable Dexterous Robotic Grasping (CVPR 2025 · similarity 0.84)
- PointZero: 3D Point Track Completion for Learning Transferable 3D Dynamics (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
