train ucsd about novelty-detection HOT 8 OPEN

Hi,

I think you are not using the correct collate function in the DataLoader.
One quick way to patch this is to reshape the input tensor before feeding the network:

x_in = x_in.view(-1, *x_in.shape[2:])

Hope this helps,
D

from novelty-detection.

@DavideA thank you，it‘s useful。 but I can't use the batch_size mentioned in the paper 。Do I need to use multiple GPUs？

from novelty-detection.

I used a single 12GB GPU.
I'd say you could reduce the batch size and give it a try.

D

from novelty-detection.

@DavideA Hi，I try to recurring your result in cifar ，but I only get AUC 0.6. could you tell me more about the detail about training。 thanks

from novelty-detection.

Hi, I have a similar question regarding input size. If my dataloader works correctly, the input data size before patching as you mentioned is [1, 690, 1, 8, 32, 32]. If I understand correctly, this first 1 is batch_size, and 8 is 8 sequential frames, and 32*32 is resized frame shape. But what is 690?

Also, the original image size is 256384, isn't 3232 too small?

Thank you in advance!

from novelty-detection.

@DavideA Hi，I try to recurring your result in cifar ，but I only get AUC 0.6. could you tell me more about the detail about training。 thanks

Hi, where did you get the training code?

from novelty-detection.

@DavideA Hi，I try to recurring your result in cifar ，but I only get AUC 0.6. could you tell me more about the detail about training。 thanks

Hi, where did you get the training code?

Same question please

from novelty-detection.

Thx for the great work! I got same question about the details about training. hoping for author uploading the training code.@DavideA
thanks again!

from novelty-detection.