search

rohban-lab / Knowledge_Distillation_AD

82 stars 46 forks source link

Training on Head CT and Brain Tumor dataset #10

Closed nono-zz closed 2 years ago

nono-zz commented 2 years ago

Hey, thanks for sharing the code, it really helps. Could you please provide code to train the model on Head and Brain dataset as well? Thank you very much.

SMSD75 commented 2 years ago

Hi,

Thank you for your interest in our research. Everything has been mentioned in the paper. You only should download the dataset and follow the instructions.

nono-zz commented 2 years ago

Hey, thanks for your reply, could you please tell me how you assign the test set from Brain Tumor and Head CT dataset?

I've downloaded these 2 datasets. The test set is composed of all of the anomalous samples and 20 normal samples, and the rest of the normals samples form the training set. I followed the same procedure as I reproduce the MVTec results, since there seems to be no specific instruction for these 2 medical datasets.

However, I didn't get the results as expected, the roc_auc value on BrainMRI and Head_CT are 0.8173 and 0.6080 respectively, I'm confused about the possible cause of my results, hope you could help me out with this!

I set the learning rate = 0.001, batch size = 64, img_size = 128, lama = 0.01.

SMSD75 commented 2 years ago

You should randomly split between the train and test samples several times. Also, for a quick suggestion, change the image size to 256. I will try to check the problem.

nono-zz commented 2 years ago

Okay, I tried splitting randomly and training for several times, the results are very close to the paper. Thank you very much!