Clone this repository and place it inside your habitat-lab/. directory (please use code from here).
- Setup episodes: Download the floorplanner episodes from here.
When training skills, add habitat.dataset.data_path=path/to/episodes.json.gz to point to the location of downloaded episodes.
There are three versions of the episodes:
| Name | Number of episodes | Scenes | Split | Object sources | Number of categories | Categories balanced? |
|---|---|---|---|---|---|---|
| cat_split_1_test_scene_40k_old.json.gz | 40k | 1 test scene | split_1 | GSO | 16 | No |
| cat_split_1_test_scene_10k.json.gz | 10k | 1 test scene | split_1 | GSO and ABO | 19 | Yes |
| cat_split_1_3_scenes_7k.json.gz | 7k | 3 train scenes | split_1 | GSO and ABO | 19 | Yes |
- Generate text embeddings:
Download clip_embeddings.pickle from here and place it inside the habitat-lab directory.
-
Download assets Download the floorplanner test scene from here and extract it inside
data.Download the "mini" floorplanner dataset with 3 training and 1 val scene from here. Download GSO and ABO objects following these instructions.
-
Train category-conditioned pick policy
./lang-rearrange-scripts/slurm_scripts/obj_rearrange/pick/cat_pick.sh
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent