This is a tools for semantic image prediction based on SAN but limit to specify semantic list. The user can input the source rgb data and the limited colormap. The tool can generate the semantic map.

Python 3.9.19

$ conda env create -n focus_san python=3.9
$ conda activate focus_san
$ pip install -r requirements.txt

Download the san_vit_large_14.pth from here and put it in the checkpoints/.

$ git clone https://github.com/Max-luo-song/FocusSAN
$ cd FocusSAN
$ python models/sanseg.py --workspace source_dir --output target_dir

Note: Both source_dir and target_dir are folders. Your RGB data are under the source_dir.

• Release the project code v1.0
• Add function: Support more kinds of semantic color list.

This project is based on https://github.com/chenhaomingbob/LabelMaker. Thanks a lot!