Gongfan Fang

> Hi @VainF, what you actually calculated with pytorch-OpCounter is MACs but not FLOPs, right? Not a very big deal but just want to get the number as accurate as...

Hi @lcr9875, I think it is easy to support different backbones by adding new entries in [network/modeling.py](https://github.com/VainF/DeepLabV3Plus-Pytorch/blob/master/network/modeling.py). But, it is a little time-consuming to train and test these models. So,...

Great! I will add Xception to this repo next week. Have a nice weekend!

Hi @lcr9875, sorry for the late feedback. Xception has been supported now!

I trained ``deeplabv3plus_xception`` on VOC2012 for only one epoch. A test sample: 

Fixed! Please try the latest version.

Hi @yiiiruu , I guess this depends on the initialization and hyperparameters. In my previous experiments, they provide similar mIoU on pascal voc. However separable_conv is slightly faster than conventional...

Hi @HPL123 , could you provide more details?

Hi there, a script for prediction was uploaded. Please refer to the latest readme for the usage.

训练的时候可以用裁剪的方式，具体裁剪大小根据你的计算资源决定。 inference阶段按照paper是需要在原始大小上测试的。