SG-DOR

Official weights for SG-DOR: Learning Scene Graphs with Direction-Conditioned Occlusion Reasoning for Pepper Plants.

Files

  • model.safetensors: complete SG-DOR model weights.
  • pytorch_model.bin: complete PyTorch state dictionary.
  • encoder.safetensors: PointNet++ MSG encoder weights with the shared_encoder. prefix removed.
  • config.json: model and preprocessing configuration.

Results

Metric Value
NDCG@3 0.851
Recall@1 0.460
Recall@3 0.818
Occlusion direction F1 0.734
Edge existence F1 0.832
Centroid MAE 0.015
Extent MAE 0.020

Links

Citation

@inproceedings{menon2026sgdor,
  title={{SG-DOR}: Learning Scene Graphs with Direction-Conditioned Occlusion Reasoning for Pepper Plants},
  author={Menon, Rohit and Mueller-Goldingen, Niklas and Pan, Sicong and Chenchani, Gokul Krishna and Bennewitz, Maren},
  booktitle={Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year={2026},
  note={Accepted}
}
Downloads last month
-
Safetensors
Model size
1.07M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Dataset used to train rohitmenon86/sg-dor

Paper for rohitmenon86/sg-dor