Final Project for Deep Learning Application II: Computer Vision
The repository contains the code to train a Conformer in a supervised manner and pre-train Conformer using SimCLR and fine-tune it on any dataset. The code also includes scripts to visualize the feature maps for conformer and perform a T-SNE analysis. Finally, I also include the code to train a Conformer using DINO.
Link to Conformer Paper and repository
Link to SimCLR Paper and repository
Link to DINO Paper and repository
For SimCLR pretraining in the SimCLR directory execute
python main.py --dataset CIFAR10
the model will be saved in the directory specified at config/config.yaml
To FineTune Conformer using a SimCLR pretrained model, specify the model path at the run.sh file and then run
sh run.sh
To visualize the feature map and attention layer output use:
python feature_maps.py -model_path model_path
