Bin Li

Yes, batch size will have a significant impact on the feature quality for SimCLR training. Please check SimCLR paper for more details. https://arxiv.org/pdf/2002.05709.pdf

1. This is just a criterion to select the best model which is the average score across AUCs, mean accuracy, and the loss value. You could change to only accuracy...

> Hi, > > How about bag_prediction in attention_map.py? Do we need to run it through a sigmoid to get the scores? When I run attention_map as it is now,...

It is chosen arbitrarily, The purpose of addition is to resemble the **Positional Encoding** in [transformers](https://arxiv.org/abs/1706.03762). You could treat this as a spatial encoding mechanism. Ideally, this number should be...

Could you check out the CSV files containing the features and labels？

> the csv seems correct... Here some screenshots of embeddings extracting using your pretrained model model_v2.pth found at https://drive.google.com/drive/folders/1_mumfTU3GJRtjfcJK_M0fWm048sYYFqi on patches extracted using 19 as threeshold: > > camelyon.csv normal143.csv...

> model_v2.pth Have you tried model_v0.pth and model_v1.pth, did they also not work?

> screenshot features using model-v0 > > screenshot features using model-v1 > > Those are very different from mine. There should not be values>10, they are all around the same...

 Does your `normal_141_42_54.jpg` look like this? My feature csv using v2 [normal_141.csv](https://github.com/binli123/dsmil-wsi/files/8756791/normal_141.csv)

It turns out that Camleyon16 consists of mixed magnifications, so by experimenting the correct configuration: `python deepzoom_tiler.py -m 1 -b 20 -d Camelyon16-pilot -v tif`