TARGO-Net

This repository hosts the released checkpoints for TARGO-Net, the model from:

TARGO and TARGO-Net: Benchmarking Target-Driven Object Grasping Under Occlusions
Accepted at International Journal of Computer Vision (IJCV), 2026.

• Project page: https://targo-benchmark.github.io/
• Paper DOI: https://doi.org/10.1007/s11263-025-02716-9
• arXiv: https://arxiv.org/abs/2407.06168

Overview

TARGO is a benchmark for target-driven 6D robotic grasping under occlusion. It evaluates how grasping performance changes as target visibility decreases, using large-scale synthetic data and real-world scenes. TARGO-Net is a transformer-based grasping model with a shape completion module, designed to remain robust as occlusion increases.

Checkpoints

File	Description
checkpoints/targonet.pt	TARGO-Net grasp prediction checkpoint.
checkpoints/adapointr.pth	AdaPoinTr target shape completion checkpoint used by the TARGO-Net pipeline.

Download

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="randing2000/TARGO-Net",
    local_dir="checkpoints_hf",
    allow_patterns=["checkpoints/targonet.pt", "checkpoints/adapointr.pth"],
)

Data

The benchmark/dataset files are not included in this model repository. Please see the project page for code, data, and benchmark details:

https://targo-benchmark.github.io/

Citation

@article{xia2026targo,
  title={TARGO and TARGO-Net: Benchmarking Target-Driven Object Grasping Under Occlusions},
  author={Xia, Yan and Ding, Ran and Qin, Ziyuan and Zhan, Guanqi and Zhou, Kaichen and Yang, Long and Dong, Hao and Cremers, Daniel},
  journal={International Journal of Computer Vision},
  year={2026},
  doi={10.1007/s11263-025-02716-9}
}

License

The model repository is released under the MIT license. Please also check the licenses of the benchmark data and any third-party assets used in your experiments.