Hierarchical Object-to-Zone Graph for Object Navigation

Sixian Zhang, Xinhang Song, Yubing Bai, Weijie Li, Yakui Chu, Shuqiang Jiang (Accepted by ICCV 2021)

ICCV 2021 Paper | Arxiv Paper | Video demo

Setup

• Clone the repository git clone https://github.com/sx-zhang/HOZ.git and move into the top-level directory cd HOZ
• Create conda environment. conda env create -f environment.yml
• Activate the environment. conda activate ng
• We provide pre-trained model of hoz and hoztpn. For evaluation and fine-tuning training, you can download them to the trained_models directory.
• Download the dataset, which refers to ECCV-VN. The offline data is discretized from AI2THOR simulator.
  The data folder should look like this

  data/ 
  └── Scene_Data/
      ├── FloorPlan1/
      │   ├── resnet18_featuremap.hdf5
      │   ├── graph.json
      │   ├── visible_object_map_1.5.json
      │   ├── det_feature_categories.hdf5
      │   ├── grid.json
      │   └── optimal_action.json
      ├── FloorPlan2/
      └── ...

  HOZ graph Construction (Updating)

  Training and Evaluation

  Train the baseline model

  python main.py --title Basemodel --model BaseModel --workers 12 -–gpu-ids 0

  Train our HOZ model

  python main.py --title HOZ --model HOZ --workers 12 -–gpu-ids 0

  Evaluate our HOZ model

  python full_eval.py --title HOZ --model HOZ --results-json HOZ.json --gpu-ids 0

  Evaluate our HOZ-TPN model

  python full_eval.py --title TPNHOZ --model MetaMemoryHOZ --results-json HOZTPN.json --gpu-ids 0

  Citing

  If you find this project useful in your research, please consider citing:

  @InProceedings{Zhang_2021_ICCV,
  author    = {Zhang, Sixian and Song, Xinhang and Bai, Yubing and Li, Weijie and Chu, Yakui and Jiang, Shuqiang},
  title     = {Hierarchical Object-to-Zone Graph for Object Navigation},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month     = {October},
  year      = {2021},
  pages     = {15130-15140}
  }