Austin model for stroke detection

To install environment:

conda create -n DSV2 python=3.9
conda activate DSV2
conda install pytorch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 pytorch-cuda=12.1 -c pytorch -c nvidia
pip install -r requirements.txt
pip install torchaudio==2.2.0
pip install marlin_pytorch

Please also refer to a previous work DeepStroke for reference. Thanks to CTA.

Main implementation:

python ds_peace_uncertain_save_sigma.py --epochs 100 --wi 1.50 --w True

wi: Test different weights in CE loss (nn.CrossEntropyLoss(weight=torch.tensor([self.wi, 1.0])))
w: when False, the second term in uncertainty loss is set to be constant 1.

other benchmarks:

We use MARLIN for visual feature extraction, and pretrained VGG model from ONE-PEACE for audio feature extraction.

To run code for uncertainty estimation experiment:

python ablate_PEACE_MAE_uncertain.py --epoch 100

We use Facexformer for visual feature extraction, and pretrained VGG model from ONE-PEACE for audio feature extraction.

To run code for experiment with facexformer and one-peace backbones:

python fxf_AST.py --epoch 100

To run code for uncertainty estimation experiment with facexformer and one-peace backbones (without visualization yet):

python fxf_AST_uncertain.py --epoch 100