Intermdiate layer matters - SSL

The official repository for "Intermediate Layers Matter in Momentum Contrastive Self Supervised Learning" paper.

1. Download the data for the experiments:

The data can be downloaded from kaggle.com. NIH chest-xray dataset: https://www.kaggle.com/nih-chest-xrays/data Breast cancer histopathology dataset: https://www.kaggle.com/paultimothymooney/breast-histopathology-images Diabetic Retinopathy dataset: https://www.kaggle.com/c/diabetic-retinopathy-detection/data

2. Training of SSL models:

To train the ssl models for moco, moco-mse and moco-btwins, please use 'train_ssl_moco.py', 'train_ssl_moco_mse.py' and 'train_ssl_moco_btwins.py' respectively. The code works for first two datasets. For the diabetic retinopathy dataset, please write a dataloader like "chest_xray_supervised.py" and a datamodule file like "chest_xray_dm.py". Import these files in 'train_ssl_moco.py', 'train_ssl_moco_mse.py' and 'train_ssl_moco_btwins.py' and make necesary changes. The same code can work for the diabetic retinopathy dataset.

3. Fine tuning the models:

To finetune the models, please use the "fine_tune_moco_chestxray.py" and "fine_tune_moco_hist.py" for NIH chest xray and Breast cancer histopathology data, respectively. For the diabetic retinopathy dataset, please write the code for fine tuning using/similar to "fine_tune_moco_chestxray.py"

4. Probing the models:

To probe the intermediate layers of the model, please use the "probing_moco_chestxray.py" and "probing_moco_hist.py" for NIH chest xray and Breast cancer histopathology data, respectively. For the diabetic retinopathy dataset, please write the code for probing the intermediate layers using/similar to "probing_moco_chestxray.py"

5. Feature reuse analysis:

To compute the feature similarity, perform the inference using your model, store the intermediate layer representations and use "CKA.py" for computing the kernel similarity with sigma = 0.8.