xiaomengyc / sg-one

SG-One: Similarity Guidance Network for One-Shot Semantic Segmentation

Hi there 👋

About Me:

⚡ Website: https://xiaomengyc.github.io/

⚡ Google Scholar: https://scholar.google.com.hk/citations?hl=en&user=BPEA2NMAAAAJ

sg-one's Issues

Installation environment

Hi, could you please provide your environment info? Like the version of Torch and the list of your site-package? Thank you.

Gain the labels

Hi,@xiaomengyc , The error happened when getting the labels based on the file of SegmentationClass.
I get the label is [127,191,223], but the catgories are 20. Then the error take place as follow:

Thank you very much!
Best regard!

more details on using the code

nice work and thanks for making it public. Could you please provide more introduction about how to use the code, including the environment, the dataset, the access to the datasets.

I tried to run the code, but it seems that some files are missed, e.g.,

NameError: name 'onemodel_v25' is not defined.

Could you kindly spare some time and refine the repo. Your effort is much appreciated.

Batch size affects the behaviour of Masked Average Pooling

In onemodel_sg-one.py line 37, the behaviour of torch.sum(pos_mask) will be different if the batch size is not 1.

It is better to write it batchsize-agnostic.

What is the file 'val_2_anns.pkl'?

In util.py, here is a function named get_anns_path in class PASCAL, I wonder what is the file that function returns? It seems there is no such a pkl file in the given path.

def get_anns_path(self, read_mode):
return osp.join(self.db_path, self.dataType + '_' + str(read_mode) + '_anns.pkl')

Load data seg bgr

Hello,

I'm looking around your code.
I've noticed that the mydataset seems to use bgr image from cv2.imread.
But you uses rgb mean and rgb pretrained pytorch vgg models.
Is there anything I missed?
I'll very appreciate if you help this out.
https://github.com/xiaomengyc/SG-One/blob/master/utils/mydataset.py#L46

Regards,
Ahyun

dat = datalayer.dequeue()

@xiaomengyc I trained the group_0_of_4 ,and achieved the path.tar files, then I used the test_frame_all.py to test, but I met with" 0%| | 0/1000 [00:00<?, ?it/s] "when debug at " for count in tqdm(range(1000)): dat = datalayer.dequeue()", it didn't work any more. Have you met with this and how can I deal with it ? and could you show some details about how to train and test the code?

AssertionError: Path does not exist: VOCdevkit2012/VOC2012/list/train.txt

From where should I download the dataset? Because I download it from http://host.robots.ox.ac.uk/pascal/VOC/voc2012/
and there is no folder named "list" and no file named "train.txt"

I am getting this error:
AssertionError: Path does not exist: /content/drive/MyDrive/sofa/data/VOCdevkit2012/VOC2012/list/train.txt

dat = datalayer.dequeue()

@xiaomengyc I trained the group_0_of_4 ,and achieved the path.tar files, then I used the test_frame_all.py to test, but I met with" 0%| | 0/1000 [00:00<?, ?it/s] "when debug at " for count in tqdm(range(1000)): dat = datalayer.dequeue()", it didn't work any more. Have you met with this and how can I deal with it ? Thank you ! Best regard!

missing train/test/val split files

The split files are missed, could you kindly upload them. Thanks.

AttributeError: 'module' object has no attribute 'onemodel_v42_7'

When I'm trying to run the scripts, it is giving the above-mentioned error.
Traceback (most recent call last):
File "train_frame.py", line 14, in
from oneshot import *
AttributeError: 'module' object has no attribute 'onemodel_v42_7'

I think that some files are missing. Can you help by providing the files?
Thanks