erichhhhho

In [here](https://github.com/meetshah1995/pytorch-semseg/blob/master/ptsemseg/models/pspnet.py#L333), we should average the overlapping prediction instead of replacing it with new psub Should replace ` pred[:, :, sx:ex, sy:ey] = psub ` with ` pred[:, :, sx:ex,...

cityscapes result

3

Hi, I saw the segmentation results are mainly in Pascal VOC, anyone tried it on Cityscapes?

Hi, when I train the VGG_Decoder from scratch for SHHB dataset, the prediction is always zero. Is it a must to start from the GCC pretrain?

Multiple outputs AttributeError

2

When certain layers have multiple outputs, there is an AttributeError: 'list' object has no attribute 'size'. Recommend adding feature to support multiple outputs.

For Camvid dataset, there are 0-11, 12 classes, isn't it? If the number of label is less, the confusion matrix seems not to be correct.

Hi, may I ask why you add [nn.Dropout2d(0.1)](https://github.com/speedinghzl/pytorch-segmentation-toolbox/blob/master/networks/pspnet.py#L74) in the PSP module, which is different from the original [implementation](https://github.com/hszhao/semseg/blob/master/model/pspnet.py#L8)

Skeleton Track: CVAT for images 1.1 annotation xml has skeleton track which has been switched off

2

### My actions before raising this issue - [x] Read/searched [the docs](https://github.com/cvat-ai/cvat/tree/master#documentation) - [x] Searched [past issues](/issues) ### Expected Behaviour For CVAT V2.2 release, the exported CVAT for images 1.1...

1. Why you select the Inception-BN as the student model here? What if you want to learn more compact and small-size student model, will this method still be applicable? 2....