How to prepare a new dataset for the method

[tiagojc]

Congratulations on your excellent work!

I'm trying to prepare a new dataset for testing the method on it but I think I'm missing some details.

Do you have detailed instructions of how to prepare it (besides the structured mentioned on data.py)?

Since now, thank you so much!

Bests

[Yunfan-Li]

Hi, what kind of data would you like to test? For image data, you could use the ImageFolder structure provided by PyTorch (https://pytorch.org/vision/stable/generated/torchvision.datasets.ImageFolder.html).

FYI, there is a similar issue in the previous repo (https://github.com/Yunfan-Li/Contrastive-Clustering/issues/10). Perhaps it would be helpful.

Best

[tiagojc]

It worked perfectly, thanks!

An additional question: in your previous work (Contrastive Clustering) there was a cluster.py which allowed to apply trained model on test images. I could not figure out which function should I use to test trained model on test samples here.

Could you help me ?

Plus, boost should take some pretrained model? I tried to load pretrained checkpoint generated from train.py but it did not work.

[Yunfan-Li]

In this implementation, the clustering performance is evaluated during the training process. The evaluation function is at line 252 in train.py: test_stats = evaluate(data_loader_val, model, device) The returned test_stats contains the clustering performance metrics.

For the second question, have you checked the "--resume" argument to make sure it corresponds to the pre-trained model? What is the error message?