SAMCL Inference

This code supports only the inference with pretrained SAM-CL model. Please visit this repository for training code.

Installation

• Clone or download the repo.
• Activate python virtual environment (venv or conda). Create virtual environment if it not available.
• Install the required Python packages with following command:

cd <path to SAMCL_Inference repo>
pip install -r requirements.txt

Checkpoint Download

Download the checkpoint from this link and place it under "ckpt" folder.

Usage Instructions:

Terminal command to run inference on stored raw thermal images:

python inference.py --datadir *<path containing .bin raw thermal file>* --outdir *<path to store segmentation masks and visualization images>* --config *<config file with input parameters>* [--gpu *gpu_number*]

Illustrative terminal commands to run inference on raw images

Inference with CPU:

python inference.py --datadir data/test/ --outdir ./out/test --config configs/AU_SAMCL.json

Inference with single GPU:

python inference.py --datadir data/test/ --outdir ./out/test --config configs/AU_SAMCL.json --gpu 0

Output

There are two outcomes that code generates in the specified outdir path:

1. PNG files having segmentation masks, with semantic regions coded as follows:

    0 - background
    1 - cheek
    2 - mouth
    3 - eyes
    4 - eyebrows
    5 - nose
   

2. JPG files (in a sub-folder with name "vis") showing overlaid visualization as the image below. These cannot be used for further processing.

📜 Citation

@inproceedings{Joshi_2022_BMVC,
    author={Jitesh N Joshi and Nadia Berthouze and Youngjun Cho},
    title={Self-adversarial Multi-scale Contrastive Learning for Semantic Segmentation of Thermal Facial Images},
    booktitle={33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
    publisher={{BMVA} Press},
    year={2022},
    url={https://bmvc2022.mpi-inf.mpg.de/0864.pdf}
}