Your Partial Conv is new in computer vision. However, if you use a ground truth image in your loss function for your model trining, your paper is worthless for image inpainting. In most cases, we only have a deteriorated image, and the the ground truth is an unknown target. about pytorch-inpainting-with-partial-conv

[naoto0804]

Although it is not my model, you are right. If one doesn't know the exact regions to be inpainted, the task is called blind image inpainting. It is much harder than the standard inpainting task, and the easiest way to solve that is by combining a region detector (or manual region identification) and an off-the-shelf standard inpainting model.

[Flying-ow]

After chatting with ChatGPT, I take back what I wrote about the paper for inpainting with a ground truth image in loss functions. Just like YOLO models, a trained inpainting model from the paper can be used for other deteriorated images, although we need a dataset with millions images and many days for model training. The algorithm presented by the paper is different from Deep Image Prior, in which a trained model can only be used for one image, but no training dataset is needed.

[Flying-ow]

After chatting with ChatGPT, I took back what I wrote about the paper for inpainting with a ground truth image in loss functions. Just like YOLO models, a trained inpainting model from the paper can be used for other deteriorated images, although we need a dataset with millions of images and many days for model training. The algorithm presented by the paper is different from Deep Image Prior, in which a trained model can only be used for one image, but no training dataset is needed.

…

________________________________ From: Naoto Inoue ***@***.***> Sent: Wednesday, January 17, 2024 4:34 PM To: naoto0804/pytorch-inpainting-with-partial-conv ***@***.***> Cc: Flying-ow ***@***.***>; Author ***@***.***> Subject: Re: [naoto0804/pytorch-inpainting-with-partial-conv] Your Partial Conv is new in computer vision. However, if you use a ground truth image in your loss function for your model trining, your paper is worthless for image inpainting. In most cases, we on... Closed #64<#64> as completed. — Reply to this email directly, view it on GitHub<#64 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A4D5CL3VI4PZM3L5PXZEYSDYPBNXJAVCNFSM6AAAAABB2JIXVOVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJRGUYTONZRGEYDQNQ>. You are receiving this because you authored the thread.Message ID: ***@***.***>