Anand Kumar, Kavinder Roghit Kanthen, Josna John
Manuscript
This is the official code for the paper "GS-TransUNet: Gaussian splatting skin lesion analysis"
This research aims to develop a more effective and accurate automated diagnostic tool for skin cancer analysis by simultaneously addressing lesion segmentation and classification tasks. Traditional methods typically handle these tasks separately, which can lead to inefficiencies and reduced accuracy. The GS-TransUNet model aims to integrate these tasks into a cohesive framework, leveraging advanced machine learning techniques to improve diagnostic precision and speed.
By combining 2D Gaussian Splatting with Transformer UNet architecture, this study seeks to enhance the consistency and accuracy of segmentation masks, which are crucial for the reliable classification of skin lesions. This integrated approach improves diagnostic accuracy and reduces the computational cost associated with separate processing stages, paving the way for real-time applications in clinical settings.
- Get models in this link: R50-ViT-B_16, ViT-B_16, ViT-L_16...
wget https://storage.googleapis.com/vit_models/imagenet21k/{MODEL_NAME}.npz &&
mkdir ../model/vit_checkpoint/imagenet21k &&
mv {MODEL_NAME}.npz ../model/vit_checkpoint/imagenet21k/{MODEL_NAME}.npz- Download the ISIC-2017 and PH-2
- Make sure to change the root paths in
dataset.pyappropriately.
Please prepare an environment with python=3.7, and then use the command "pip install -r requirements.txt" for the dependencies.
- Run the train script
CUDA_VISIBLE_DEVICES=0 python train.py --xp_name gaussOnce training is done the script automatically runs test.