pramuperera / deeponeclass Goto Github PK

Dear authors,
I want to train a model that can detect if my image is a lottery or not. I'm wondering how my data folder look like? And how I should change config to train my own dataset without imagenet validation?
Thanks for your help!

Hi everyone, I would like to ask if anyone has an implementation for this method in PyTorch. If yes, please let me know in the comment. This would be extremely helpful to me. Thank you so much!

Hello, thank you for your work! I met a problem when reading the paper. When the compactness loss is calculated, each sample for the loss function has a dimension of n*k, what the k means about?
Thank you.

Hello PramuPerera,
We are a team of student in France trying to get your repository working for abnormal image detection,
but there is a dead link (the author changed his website) : http://paul.rutgers.edu/~babaks/abnormality_detection.html
Do you know where we could find the ressources ?
Thank you !
Valink

I think the implementation of the distance_layer.py has some problem,

the reshape function has some not necessary process, and the code in backward function is not true as in your paper.

Regarding these 2 images.

My understanding is that compactness loss should only be calculated from the target batch. And descriptiveness loss should only be calculated from the reference batch.

Am I correct to understand that the weights reference network and secondary network are identical across all gradient updates?

In other words, for the W in the equation, does that refer to the weights representing the combined weights from the reference network and secondary network, or does that mean the reference network's weights and secondary network's weights are exactly the same set of weights?

If is is the former, then it fine to have 2 network stacks that are being trained from the different loss functions, and their weights will deviate, with the secondary network learning the compactness of the batch, while the reference network being trained to do multi-class classification.

If it is the latter, then I am confused as to how the weights are made identical without freezing entirely both networks.

Hi, thank you for your fantastic work! I think you did augmentation on the normal object dataset. I couldn't find details in your paper. What augmentations did you do? Could you please share the Normal Object Dataset you use for training/testing after augmentation so that I can get similar results as presented in your paper. Thank you very much.

Hi,
I have read your paper and I think it is a very good way to combine both compactness and descriptiveness, but I got a little confused about the Reference Network and Secondary Network, you mentioned in your paper in section 3.4 that 'weights between two networks are tied together forcing them to be identical.', does that means the weights of the two networks are identical during training ( for both g(·) and hc(·) )?
Thank you for your time and have a good day!

Hi, Thanks for your works.
Do you think the type of reference dataset needs to be similar to the target one? I am implementing Pytorch on my custom dataset (scene dataset), Do I need to find another scene dataset to be the reference? Because Imagenet dataset ( original paper) focuses on the specific object which is similar to the target object (chair)

I think the idea in FaceNet solves the same problem. Have you compared your method to triplet loss?

Sorry to bother you, but how to train on a new dataset? thank you.