liheyoung/miningfss

[ICCV 2021 Oral] Mining Latent Classes for Few-shot Segmentation

miningfss's Issues

Offline Annotation and Mining Embedding

Hello, thank you for your great contribution!
I have a question: Can you provide the code of Offline Annotation and Mining Embedding? It seems like the repository doesn't include them.

about coco trained model

It's a great job. Will you publish coco trained model?

pseudo loss and offline annotation

Thanks for sharing the code. i wonder where the code is about your offline annotation? and i do not see the pseudo loss during the training period.

Prototype Rectification

Does this code contains any prototype rectification process mentioned in the paper ? I do not observe such moving average operation during episodic training. I think you share training without rectification. Is it true ?

Data replacement

Hello, I realized the effect of the paper. I changed the experimental data into my own data. Classification category: 1, 2, 3. The label image is as follows

For my above classification, how to change fewshow Py code, the screenshot is as follows

Thank you for your guidance

How to begin training in COCO2014 dataset?

Thanks for your perfect work!
The problem like the title, I want to know how to begin training the coco2014 dataset.

Source code for reproducing the results

Hi Authors,

Thank you very much for your great work. Could you release the code of your method, rather than just the baseline?

the Pascal JPEGImages dataset can't be downloaded

thank you for the great work,I have some problem with the code,does the initial epoch is 20? I have runned the code ,but it seemed worked not so well as the paper says

offline Annotation

I only saw the code of Loss_gt branch, but did not see the code about offline annotation operation.
Please supplement code as soon as possible。

About Trained Models

Hello, thank you for your great contribution!
I have a question: trained model is pretrained backbone? It seems like each trained model which you provided is not the best model.

Will you please produce the source code?

VOC

how to initialise global background prototype?

Hi,

Can I how to initialise the global background prototype P_{bg}^{global} in eqn. 6 in the paper? I didn't see it in the code. Thanks!

About pretrained embedding network on offline annotation

I have a question in the offline annotation part.
It is mentioned that it used the pretrained embedding network, does the pretrained network mean the imagenet pretrained weight? If not, may I know how did you learn the model?