search

Reyhanehne / CVF-SID_PyTorch

Official implementation of the paper "CVF-SID: Cyclic multi-Variate Function for Self-Supervised Image Denoising by Disentangling Noise from Image" (CVPR 2022)
71 stars 11 forks source link

Question of the performance on SIDD validation dataset #8

Open alwaysuu opened 1 year ago

alwaysuu commented 1 year ago

Thanks for sharing the source code and the pre-trained model on SIDD validation dataset. I tested the provided pre-trained model on SIDD validation dataset and only got 34.14/0.855 (SPNR/SSIM). image

Moreover, I tried to train the denoiser ( CVF-SID (S) ) on SIDD validation dataset directly, and I can only get 34.05/0.847.

I am using the same versions of python and pytorch as yours. I would like to know if I missed something important.

Reyhanehne commented 1 year ago

Please make sure that you trained the pre-trained model on the SIDD-Medium dataset not the SIDD-Small dataset.

alwaysuu commented 1 year ago

Thanks for the reply. First, I would like to know why I cannot get the performance as yours by using the provided pre-trained model on SIDD validation dataset. Do I need to continue training the pre-trained model to get better performance? Second, when I follow the procedure of CVF-SID (S) and directly train the model on the testset (SIDD validation), I also cannot get the performance as yours. I think the way you mentioned "make sure that you trained the pre-trained model on the SIDD-Medium dataset" should be CVF-SID (T)? image

lulu-165 commented 1 year ago

Hello, may I ask what the first and second dimensions in the data set respectively mean? image

alwaysuu commented 1 year ago

Hello, it means that the validation set has 40 images with 32 blocks, each of which has a size (256,256,3)

Reyhanehne commented 1 year ago

Hello, Please refer to Section 4.1 of the main paper.
"Validation and benchmark splits are used, each containing 32 patches of size 256 × 256 from 40 images".

lulu-165 commented 1 year ago

ok,please  give me your number.

------------------ 原始邮件 ------------------ 发件人: "Reyhanehne/CVF-SID_PyTorch" @.>; 发送时间: 2023年5月1日(星期一) 中午12:56 @.>; @.**@.>; 主题: Re: [Reyhanehne/CVF-SID_PyTorch] Question of the performance on SIDD validation dataset (Issue #8)

你好，请问数据集中的第一维度和第二维度分别是什么意思？

Can I have your WeChat account? I am also working on this project

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

DiffDynamo commented 1 year ago

Thanks for sharing the source code and the pre-trained model on SIDD validation dataset. I tested the provided pre-trained model on SIDD validation dataset and only got 34.14/0.855 (SPNR/SSIM). image

Moreover, I tried to train the denoiser ( CVF-SID (S) ) on SIDD validation dataset directly, and I can only get 34.05/0.847.

I am using the same versions of python and pytorch as yours. I would like to know if I missed something important. Can I ask if your issue has been resolved? Have you been able to replicate the results from the paper?

DiffDynamo commented 1 year ago

好啊，我的微信二维码在附件中。

------------------ 原始邮件 ------------------ 发件人: "Reyhanehne/CVF-SID_PyTorch" @.>; 发送时间: 2023年5月20日(星期六) 下午4:36 @.>; @.**@.>; 主题: Re: [Reyhanehne/CVF-SID_PyTorch] Question of the performance on SIDD validation dataset (Issue #8)

可以加微信聊，我也只是复制出结果，还没换数据集

---Original--- From: @.> Date: Sat, May 20, 2023 16:32 PM To: @.>; Cc: @.**@.>; Subject: Re: [Reyhanehne/CVF-SID_PyTorch] Question of the performance on SIDDvalidation dataset (Issue #8)

Thanks for sharing the source code and the pre-trained model on SIDD validation dataset. I tested the provided pre-trained model on SIDD validation dataset and only got 34.14/0.855 (SPNR/SSIM).

Moreover, I tried to train the denoiser ( CVF-SID (S) ) on SIDD validation dataset directly, and I can only get 34.05/0.847.

I am using the same versions of python and pytorch as yours. I would like to know if I missed something important. Can I ask if your issue has been resolved? Have you been able to replicate the results from the paper?

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