How to test your model on my dataset, png format

[stonecropa]

Hello, I want to use your pre-trained model to denoise my oct data set, but I see that your code is all in nii format, and my images are in png format. I don’t know how to modify it. to test directly on your pre-trained model. I would be grateful if you could answer me.

[DeweiHu]

Hi, the image format doesn't matter as long as you convert it to pytorch tensor with shape [b, c, h, w] and the intensity normalized to range [1, 3].

[stonecropa]

Thank you. I know how to modify it, if I have any questions, I will come and ask you. Thank you very much.

[stonecropa]

Hello, it may be because of the noise in my image. There is still a certain gap between the clear image generated by the pre-trained model and the clear image in your paper.The first image is the original image, and the others are denoised images.(it is t=40, 45 50,55,60,65 ). Do I need to retrain?

[DeweiHu]

Yes, to acquire better performance, you probably need to fine-tune the model on your own dataset considering the inherent distribution shift. And you can print the whole spectrum of t to pick the optimal value.

…

________________________________ From: stonecropa ***@***.***> Sent: Wednesday, October 25, 2023 1:26 AM To: DeweiHu/OCT_DDPM ***@***.***> Cc: Hu, Dewei ***@***.***>; Comment ***@***.***> Subject: Re: [DeweiHu/OCT_DDPM] How to test your model on my dataset, png format (Issue #10) Hello, it may be because of the noise in my image. There is still a certain gap between the clear image generated by the pre-trained model and the clear image in your paper.The first image is the original image, and the others are denoised images.(it is t=40, 45 50,55,60,65 ). Do I need to retrain? [1]<https://user-images.githubusercontent.com/114655828/277893203-f3c2ad04-4d2f-4001-9f87-afd300a095bb.png> [image_40]<https://user-images.githubusercontent.com/114655828/277893668-de4694ba-232e-4520-8bbf-e6ffa1b6d961.png> [image_45]<https://user-images.githubusercontent.com/114655828/277893478-b9e18be9-d407-4d59-b37a-6fd5c49b9fc8.png>  [image_55]<https://user-images.githubusercontent.com/114655828/277893611-70a0ea56-be80-4c1a-be81-1191b003bbb6.png> [image_60]<https://user-images.githubusercontent.com/114655828/277893739-5c7cefa0-4706-4180-9f73-e6467c62d9f1.png> [image_65]<https://user-images.githubusercontent.com/114655828/277893759-dc970f5c-0581-4357-bb37-8927f037fd81.png> — Reply to this email directly, view it on GitHub<#10 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AL6WZQBWLB2UTT6RQPWXRQDYBCWH3AVCNFSM6AAAAAA6K54JCWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZYGU4TGOJTHE>. You are receiving this because you commented.Message ID: ***@***.***>

[stonecropa]

@DeweiHu
Hello, how can I fine-tune your pre-trained model, and how can I print out the frequency spectrum of t? I would be very grateful if you could tell me the detailed steps.

[san-kou7]

谢谢。我知道如何修改它，如果我有任何问题，我会来问你。谢谢。

Hello, I would like to know how you modified the code or processed images. My own images are not in nii format. If you could answer me, I would greatly appreciate it @stonecropa

[gguyinging]

您好，如何微调您的预训练模型，以及如何打印出 t 的频谱？如果您能告诉我详细的步骤，我将不胜感激。
您好，我想知道您是如何修改代码或处理图像的，如果您能回答我，我将不胜感激