Thanks for your great work! Can you share how to implemented t-SNE visualization of the high-level embeddings from the last layer of the image encoder on CheXpert 5x200？

[Dramaeyp]

[QtacierP]

Thanks for your attention. We utilize sklearn.manifold.TSNE to implement the T-SNE process. We use the last layer's output from the image encoder as the features. Maybe this question is a good example.