milesial / Pytorch-UNet

PyTorch implementation of the U-Net for image semantic segmentation with high quality images
GNU General Public License v3.0
9.01k stars 2.47k forks source link

where is the skip-connect? #456

Open wangwxr opened 1 year ago

wangwxr commented 1 year ago

where is the skip-connect???? it's missng in the code

WenBingo commented 1 year ago

hi , I want to know how to down the dataset. I run the code of "bash scripts/download_data.sh", but need the “Kaggle username”and " Kaggle username". Please, i don't know how to solve this.

wangwxr commented 1 year ago

下个git

------------------ 原始邮件 ------------------ 发件人: "milesial/Pytorch-UNet" @.>; 发送时间: 2023年8月30日(星期三) 下午4:16 @.>; 抄送: "Ghost @.**@.>; 主题: Re: [milesial/Pytorch-UNet] where is the skip-connect? (Issue #456)

hi , I want to know how to down the dataset. I run the code of "bash scripts/download_data.sh", but need the “Kaggle username”and " Kaggle username". Please, i don't know how to solve this.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

WenBingo commented 1 year ago

嗯,就是在git上运行的 ,随后就提示Kaggle username输入, Bin@DESKTOP-DLQF623 MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username:

Bin@DESKTOP-DLQF623 MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username:

wangwxr commented 1 year ago

输入kaggle的用户名和密码

------------------ 原始邮件 ------------------ 发件人: "milesial/Pytorch-UNet" @.>; 发送时间: 2023年8月30日(星期三) 下午4:22 @.>; 抄送: "Ghost @.**@.>; 主题: Re: [milesial/Pytorch-UNet] where is the skip-connect? (Issue #456)

嗯,就是在git上运行的 ,随后就提示Kaggle username输入, @.*** MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username:

@.*** MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username:

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

WenBingo commented 1 year ago

Bin@DESKTOP-DLQF623 MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Requirement already satisfied: kaggle in d:\anaconda\lib\site-packages (1.5.16) Requirement already satisfied: requests in d:\anaconda\lib\site-packages (from kaggle) (2.28.1) Requirement already satisfied: certifi in d:\anaconda\lib\site-packages (from kaggle) (2022.12.7) Requirement already satisfied: python-slugify in d:\anaconda\lib\site-packages (from kaggle) (5.0.2) Requirement already satisfied: bleach in d:\anaconda\lib\site-packages (from kaggle) (4.1.0) Requirement already satisfied: six>=1.10 in d:\anaconda\lib\site-packages (from kaggle) (1.16.0) Requirement already satisfied: python-dateutil in d:\anaconda\lib\site-packages (from kaggle) (2.8.2) Requirement already satisfied: urllib3 in d:\anaconda\lib\site-packages (from kaggle) (1.26.14) Requirement already satisfied: tqdm in d:\anaconda\lib\site-packages (from kaggle) (4.64.1) Requirement already satisfied: packaging in d:\anaconda\lib\site-packages (from bleach->kaggle) (22.0) Requirement already satisfied: webencodings in d:\anaconda\lib\site-packages (from bleach->kaggle) (0.5.1) Requirement already satisfied: text-unidecode>=1.3 in d:\anaconda\lib\site-packages (from python-slugify->kaggle) (1.3) Requirement already satisfied: charset-normalizer<3,>=2 in d:\anaconda\lib\site-packages (from requests->kaggle) (2.0.4) Requirement already satisfied: idna<4,>=2.5 in d:\anaconda\lib\site-packages (from requests->kaggle) (3.4) Requirement already satisfied: colorama in d:\anaconda\lib\site-packages (from tqdm->kaggle) (0.4.6) 401 - Unauthorized unzip: cannot find or open train_hq.zip, train_hq.zip.zip or train_hq.zip.ZIP. mv: cannot stat 'train_hq/': No such file or directory rm: cannot remove 'train_hq': No such file or directory rm: cannot remove 'train_hq.zip': No such file or directory 401 - Unauthorized unzip: cannot find or open train_masks.zip, train_masks.zip.zip or train_masks.zip.ZIP. mv: cannot stat 'train_masks/': No such file or directory rm: cannot remove 'train_masks': No such file or directory rm: cannot remove 'train_masks.zip': No such file or directory

不好意思,这些报错了....

WenBingo commented 1 year ago

想问下这个数据集是自己准备吗?我一直认为是作者给了运行数据集,然后运行这个是下载数据集的

WenBingo commented 1 year ago

输入kaggle的用户名和密码 ------------------ 原始邮件 ------------------ 发件人: "milesial/Pytorch-UNet" @.>; 发送时间: 2023年8月30日(星期三) 下午4:22 @.>; 抄送: "Ghost @.**@.>; 主题: Re: [milesial/Pytorch-UNet] where is the skip-connect? (Issue #456) 嗯,就是在git上运行的 ,随后就提示Kaggle username输入, @. MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username: @. MINGW64 /e/Pytorch-UNet-master/scripts $ bash download_data.sh Kaggle username: — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

哥,作者的数据集你能用百度网盘、阿里云盘分享一下吗?我这边Git一直下载报错 拜托

naguileraleal commented 1 year ago

Hello. Here is the skip connect https://github.com/milesial/Pytorch-UNet/blob/2f62e6b1c8e98022a6418d31a76f6abd800e5ae7/unet/unet_parts.py#L67

wangwxr commented 12 months ago

OK Thank you. I have some ideas want to communicate with you. Do you think sgd is better than other optimizers in U-net?

------------------ 原始邮件 ------------------ 发件人: "milesial/Pytorch-UNet" @.>; 发送时间: 2023年9月25日(星期一) 凌晨2:57 @.>; 抄送: "Ghost @.**@.>; 主题: Re: [milesial/Pytorch-UNet] where is the skip-connect? (Issue #456)

Hello. Here is the skip connect https://github.com/milesial/Pytorch-UNet/blob/2f62e6b1c8e98022a6418d31a76f6abd800e5ae7/unet/unet_parts.py#L67

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

MartinSchmauch commented 10 months ago

where is the skip-connect???? it's missng in the code

I was searching for it quite a while as well. The skip connection is actually included in the forward method of the UP class in unit_parts.py. That is why the up methods in the undet_model.py also get 2 parameters instead of 1 as for the down

Also there is some padding added to the upsampled tensor for dimension matching as the corresponding tensor from the downsampling path is bigger as you can see from the unet architecture. although I am not quite sure if this is the right way to go.. in the original paper on unets the tensor from the downsampling path was cropped rather than to pad the tensor from the upsampling path