Code Question

[AmingWu]

Dear Authors,

For this error, could I only delete the code of self.writer.close()?

[zaiquanyang]

Dear Authors,

For this error, could I only delete the code of self.writer.close()?

I also meet this case and it seems does not affect the normal training.
Besides, I have other concerns.

1. What makes me confused is the split of VOC2012. When I use the split of train 10582 and test 1449, under ZS3 settings, I seem to obtain better results (2 unseen classes, the hIoU achieves 60%). In fact, I do not change the detail settings.
2. The metric of the work seems to include the background. I am not sure which it is reasonable.
3. the settings of ZS3 task are so chaotic .....

[aoyanl]

Dear Authors,
For this error, could I only delete the code of self.writer.close()?

I also meet this case and it seems does not affect the normal training. Besides, I have other concerns.

1. What makes me confused is the split of VOC2012. When I use the split of train 10582 and test 1449, under ZS3 settings, I seem to obtain better results (2 unseen classes, the hIoU achieves 60%). In fact, I do not change the detail settings.
2. The metric of the work seems to include the background. I am not sure which it is reasonable.
3. the settings of ZS3 task are so chaotic .....

I also confused of the results. I got the worse results( 2 unseen classes, the hiou achieves 48.3%, 53.1% reported in paper ) , and i didn't change any trainning setting.

[Mamduh-k]

Dear Authors,
For this error, could I only delete the code of self.writer.close()?

I also meet this case and it seems does not affect the normal training. Besides, I have other concerns.

1. What makes me confused is the split of VOC2012. When I use the split of train 10582 and test 1449, under ZS3 settings, I seem to obtain better results (2 unseen classes, the hIoU achieves 60%). In fact, I do not change the detail settings.
2. The metric of the work seems to include the background. I am not sure which it is reasonable.
3. the settings of ZS3 task are so chaotic .....

I also confused of the results. I got the worse results( 2 unseen classes, the hiou achieves 48.3%, 53.1% reported in paper ) , and i didn't change any trainning setting.

I also cannot use the author's code to achieve the results in the paper.