DeepLab-v3-plus-cityscapes

DeepLab-v3-plus-cityscapes copied to clipboard

About BatchNormalization

8

hi by running your project,I can reach high mIoU,but when the Batchnormlization is setting all SYNC in ResNet,the mIoU can just reach 70%,why is that?

RainPot

How to label my own image

1

Thank you for your coding. I want to use my own images to train the model. So when I label my images, I just need to label them like "xxxxgtFine_labelIds.png...

huidemie

PermissionError: [Errno 13] Permission denied: '/tmp/torch_extensions/inplace_abn'

2

 I am a beginner, I would like to ask you how to solve this mistake?

syp2ysy

It is difficult to download from baidu, would it be possible for you to upload the pre-trained model somewhere else, like google drive. Thank You

ManuMBhat

different results with same pth

1

I get different results from the same weight. No parameter change in the whole case. I got miou 80 in the end of the training after the weight was saved....

ywangeq

leedoge

evluate.py

1

Hi, I noticed you updated your evluate.py. But I think your **previous** implementation is right, since if you use the evaluation code from https://github.com/mcordts/cityscapesScripts, you can get a mIou ~78.x...

csBob123

Did you run the model on 8 GPUs or more? I cannot get a better result with more GPU, in fact it is worse.

LetsGoFir

ConvBNReLU

2

There is no ReLU for ConvBNReLU in deeplabv3plus.py. Is that right? @CoinCheung

LiQiufu

train loss

28

what about your train loss at last , 0.1xxx or0.01xxxx?

seeyouagain111