└── generation
│── SLIC
│── SAM
│── OpenSeeD
│── X-Decoder
│── SEEM
│── FastSAM
└── Semantic-SAM
- [2023.07] - We provide the code for generating semantic superpixels through SLIC, SAM, and SEEM. Meanwhile, we include the code for building correspondence between semantic superpixels and semantic superpoints on the nuScenes dataset. Our next update aims to integrate more VFMs, such as OpenSeeD, X-Decoder, and FastSAM.
-
Kindly follow the official guideline of SAM and SEEM for the installation details.
-
Note that nuScenes devkit is required in order to generate semantic superpixels on the nuScenes dataset.
pip install nuscenes-devkit
-
Semantic superpixel generation using SLIC:
cd generation/SLIC python superpixel_generation.py
-
Semantic superpixel generation using SAM:
cd generation/SAM python superpixel_generation.py
-
Semantic superpixel generation using SEEM:
cd generation/SEEM/demo_code python superpixel_generation.py
- The following code builds the 2D-3D correspondence between the semantic superpixel and semantic superpoint:
cd generation python superpixel2superpoint.py