helioszhao / ncdss Goto Github PK

Hi,
thanks for your source code, but I have a question
You use base_train_fold0.txt to train a supervised segmentation model f_b in stage1. Then use f_b to generate predicts for file in fold0/novel.txt. However, the part of the filenames are in base_train_fold0.txt. I'd like to ask if there has some error.

Hope to get your reply.

hello~

May I ask how you obtain the saliency maps?

I only see the saliency maps of VOC. There are not saliency maps of Cityscapes, nor the saliency model. Could you please provide them, or where can I find them? Thanks!

what are the modifications i need to make while using my binary custom dataset, because encountered error when i replaced with my dataset, this is the error message. File "/home/mohammed/NCDSS/utils/common_config.py", line 110, in get_train_dataset
dataset = VOC12_Base(root=p['data_root'], split=p['train_db_kwargs']['split'], transform=transform, novel_fold=p['fold'])
File "/home/mohammed/NCDSS/data/dataloaders/pascal_voc.py", line 226, in init
assert os.path.isfile(_image)
AssertionError

Hello, thanks for your enlightening work!

I am deeply interested in your COCO experiment.
Can you provide the trained models for COCO task?

Thanks!

I have a bit of a doubt about this code of base_train.py. I think the location of some codes should be reversed, right?

for name, param in model.named_parameters():
    for ft_layer in p['ft_layer']:
        if name.startswith(ft_layer):

            param.requires_grad = True
            if name.startswith('decoder'):
                print('Add {} in decoder params'.format(name))
                decoder_params.append(param)
            else:
                print('Add {} in classifier params'.format(name))
                classifier_params.append(param)
            
            break
        else:
            param.requires_grad = False

Could you please tell the answer？thank you very much!!!

RuntimeError: CUDA out of memory. Tried to allocate 512.00 MiB (GPU 0; 10.76 GiB total capacity; 9.10 GiB already allocated; 101.44 MiB free; 9.35 GiB reserved in total by PyTorch)
i received this error, even when i made my batch size 2.

Hello, and thank you for the work and for making it publicly available!

I'm trying to reproduce your results in the paper by downloading the datasets and running the eval.sh script.

What I'm getting at the moment for the four folds is:

69.4, 65.32, 61.3, 65.15

Whereas in the paper, the results reported are:

69.79, 60.11, 56.28, 50.18

Could a gain of 15 points on the last fold be possible?

Many thanks for considering my request.
Alessandro

Hi, thanks for the good research and code release.

Can you provide the trained models of stage-1?

Thank you!

Sorry, I did not find the training code on the COCO-20 dataset. Can you share it with me? Thank you!

I encountered some mistakes when I started training and would like to ask you how to solve them. Looking forward to your reply！

Hi, thanks for your excellent work!

May I ask how the data/data_split/* are produced? Thanks!