search

nazmul-karim170 / UNICON

[CVPR'22] Official Implementation of the CVPR 2022 paper "UNICON: Combating Label Noise Through Uniform Selection and Contrastive Learning"
MIT License
60 stars 15 forks source link

Implementation of dataloader code #8

Closed bonggeon closed 1 year ago

bonggeon commented 1 year ago

Hi. thanks for interesting paper and available code.

I have a question for dataloader code.

I think this code should be added in dataloader code. pred_idx = pred_idx[:size_pred]

for example, in dataloader_cifar.py line 169 self.train_data = train_data[pred_idx]

In github code, last few pred_idx are 0.

This should be applied to all the dataloader code.

If I misunderstood something, please let me know

thank you.

nazmul-karim170 commented 1 year ago

I am sorry I did not get your question.

Nazmul

On Fri, Jan 6, 2023 at 05:06 bonggeon @.***> wrote:

Hi. thanks for interesting paper and available code.

I have a question for dataloader code.

I think this code should be added in dataloader code. pred_idx = pred_idx[:size_pred]

for example, in dataloader_cifar.py line 169 self.train_data = train_data[pred_idx]

In github code, last few pred_idx are 0.

This should be applied to all the dataloader code.

If I misunderstood something, please let me know

thank you.

— Reply to this email directly, view it on GitHub https://github.com/nazmul-karim170/UNICON-Noisy-Label/issues/8, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF24QPPHYYG3MT5ZOVLXUJ3WQ7VBHANCNFSM6AAAAAATS5IZPQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

bonggeon commented 1 year ago

In dataloader_cifar.py line 139 pred_idx = np.zeros(int(self.sample_ratio*num_sample))

but I think size_pred < int(self.sample_ratio*num_sample)

so, pred_idx[size_pred:] are all zeros

this affects line 169 self.train_data = train_data[pred_idx]

If I have misunderstanding, I'm sorry but please think about it and let me know.

nazmul-karim170 commented 1 year ago

You can use append to a list instead of array initialization.

Like,

pred_idx = []

pred_idx.append()

On Fri, Jan 6, 2023 at 12:21 bonggeon @.***> wrote:

In dataloader_cifar.py line 139 pred_idx = np.zeros(int(self.sample_ratio*num_sample))

but I think size_pred < int(self.sample_ratio*num_sample)

so, pred_idx[size_pred:] are all zeros

this affects line 169 self.train_data = train_data[pred_idx]

If I have misunderstanding, I'm sorry but please think about it and let me know.

— Reply to this email directly, view it on GitHub https://github.com/nazmul-karim170/UNICON-Noisy-Label/issues/8#issuecomment-1373916564, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF24QPOMIGTPKH5XYXSTCSTWRBIBJANCNFSM6AAAAAATS5IZPQ . You are receiving this because you commented.Message ID: @.***>

bonggeon commented 1 year ago

I got it. thanks!