SIGN

Project website

1. Install

1.1 Install habitat-lab

cd SIGN/

# create conda env
conda create -n sign python=3.8
conda activate sign

# install habitat-sim
conda install habitat-sim=0.2.2 withbullet headless -c conda-forge -c aihabitat

# install pytorch
pip3 install torch torchvision torchaudio

# clone habitat-lab code
git clone https://github.com/facebookresearch/habitat-lab.git
cd habitat-lab
git checkout 1f7cfbdd3debc825f1f2fd4b9e1a8d6d4bc9bfc7
pip install -e habitat-lab 
pip install -e habitat-baselines

1.2 Install other requirements

cd ..
pip install -r requirements.txt

2. Download Dataset

Dataset preparation refers to https://github.com/XinyuSun/FGPrompt

<pre> data ├── datasets │ └── imagenav │ ├── gibson │ │ ├── v1 │ │ ├── train │ │ └── val │ ├── hm3d │ │ ├── v1 │ │ ├── val_easy │ │ ├── val_hard │ │ └── val_medium │ └── mp3d │ ├── v1 │ ├── test_easy │ ├── test_hard │ └── test_medium └── scene_datasets ├── gibson │ ├── xx.glb │ └── xx.navmesh ├── hm3d │ ├── scene_id │ ├── xx.glb │ └── xx.navmesh └── mp3d ├── scene_id ├── xx.glb └── xx.navmesh </pre>

3. Run ImageNav Training

python run_sim.py --overwrite \
--exp-config exp_config/ppo_imagenav_gibson.yaml,policy,reward,sensors \
--run-type train --model-dir results/imagenav/training_log

4. Run Evaluation

python run_sim.py \
--exp-config exp_config/ppo_imagenav_gibson.yaml,policy,reward,sensors,eval \
--run-type eval --model-dir ckpt \
habitat_baselines.eval_ckpt_path_dir ckpt/nav_model.pth

5. Cross-domain Evaluation

difficulty= [val_easy, val_hard, val_medium]

python run_sim.py \
--exp-config exp_config/ppo_imagenav_hm3d.yaml,policy,reward,sensors,eval \
--run-type eval --model-dir ckpt \
habitat_baselines.eval_ckpt_path_dir ckpt/nav_model.pth \
habitat_baselines.eval.split val_easy

difficulty= [test_easy, test_hard, test_medium]

python run_sim.py \
--exp-config exp_config/ppo_imagenav_mp3d.yaml,policy,reward,sensors,eval \
--run-type eval --model-dir ckpt \
habitat_baselines.eval_ckpt_path_dir ckpt/nav_model.pth \
habitat_baselines.eval.split test_easy