ResUnet

ResUnet copied to clipboard

when I draw your net I find that your res block does not correspond with the usual residual structure in my memory

1

there is no need for conv_skip(x) just conv_block(x)+x,your step is like neither resnet nor resnext. If I understand wrong and hope to reply, thank you.

handsn

Hello, thank you for your work! May I ask can you provide some pretrained checkpoints for your ResUnet and ResUnet-plus-plus, as I find training the model from scratch very difficult...

yunxdai

Can you please explain how to test the model after training.

shivam-7500

模型准确率问题

1

您好，首先感谢您的分享，我试着用您的网络模型做一个分割，为什么在训练时准确率一直很低，上不去呢？请问是否能告知一下原因

libo-coder

I want to change the data to RGB building dataset, how do I edit the code?

BigBigOranges

rishikksh20

https://github.com/rishikksh20/ResUnet/blob/01566c9ca77184ec7ecbd21ddb0681b5941e63f4/core/modules.py#L5 Hi, why is this order (BN,Relu,Conv,BN,..) used in the ResBlock and not the order(Conv,BN,Relu,Conv,...) as in the paper of the ResNet ? Many thanks and best regards

nkoser

为什么我的数据集导不进去

1

(resunet) root@8d8322c93c35:/project/resunet# python preprocess.py --config "configs/default.yaml" --train training --valid testing Length of image : 0 Length of mask : 0 Cropping Training images: 0it [00:00, ?it/s] Cropping Validation images: 0it...

sujikun

JOjingxuan

This repo has almost 300 stars but does *NOT* have residual connections

2

Author wrote: ``` def forward(self, x): return self.conv_block(x) + self.conv_skip(x) ``` Which will function like a grouped convolution. A skip would be: ``` def forward(self, x): return self.conv_block(x) + x...

MatthewBM