"After using your TLC module, the restored image has a blocking effect. Do you know how to alleviate it?"

[Redamancy0222]

[achusky]

Hi, could you please provide more information about your case? For example, which particular model was utilized for which tasks? Also, could you kindly supply an example input and the corresponding restored image?

This will help me a lot to understand the unpleasant results. Thanks.

[Redamancy0222]

First of all, thank you for your reply. "I tried to graft your TLC module onto a deblurring model that I was researching myself, and then some results were prone to significant chunking effects, especially in real photos taken by myself. I set the TLC sizes to 256 and 32.". My model uses Instance Normalization.

南理工郭政 @.***

 

------------------ 原始邮件 ------------------ 发件人: "megvii-research/TLC" @.>; 发送时间: 2023年3月26日(星期天) 中午1:10 @.>; @.@.>; 主题: Re: [megvii-research/TLC] "After using your TLC module, the restored image has a blocking effect. Do you know how to alleviate it?" (Issue #16)

Hi, could you please provide more information about your case? For example, which particular model was utilized for which tasks? Also, could you kindly supply an example input and the corresponding restored image?

This will help me a lot to understand the unpleasant results. Thanks.

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

[Redamancy0222]

BaiduYun：https://pan.baidu.com/s/1Xk3SY2h4orkLN4o7sHGKAg?pwd=qdxg Keyword：qdxg

[achusky]

Hi, thank you for providing the information.

Regarding your setting, "I set the TLC sizes to 256 and 32," I have a question. Why are there two numbers for the size of TLC? Does it mean $256\times32$? Or do you crop the original input image with size $256\times256$ and use a stride or overlapping size of $32$?

Also, I tested your example inputs with HINet-Local (HINet with our TLC), and I did not find any significant chunking effects in the restored images. Our TLC converts the (global) Instance Normalization in HINet into a local Instance Normalization. You can try HINet-Local in Google Colab. Please let me know if I missed anything.