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
• Install the dependencies. pip install -r requirements.txt
• 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.
  Your 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

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