From 3215c5a510cb0d034c4fe6a1c69ee49d822ead5a Mon Sep 17 00:00:00 2001 From: Pietro Morerio Date: Mon, 4 Mar 2024 17:20:17 +0100 Subject: [PATCH] Remove spurious files --- updates.md | 28 ---------------------------- 1 file changed, 28 deletions(-) delete mode 100644 updates.md diff --git a/updates.md b/updates.md deleted file mode 100644 index 2fbad6b..0000000 --- a/updates.md +++ /dev/null @@ -1,28 +0,0 @@ -# Puzzle Diffusion - -## Supporting idea -Diffusion models solve problems of recovering $x_0$ by "denoising" a noisy data $x_T$ with a chain process such that $x_0=p(x_T)\prod^T{p(x_{t-1}|x_t)}$ - -For images, typically $x_0$ is the image, and $x_T$ is noise sampled from $\mathcal{N}(0,1)$. - -In our case we want to recover position of 2D patches of in image. So $p_\theta(pos_t | pos_{t-1},rgb)$. - ---- -## TODO -- [x] Add DDIM, with skippable steps -- [ ] Test rotations -- [ ] Test missing pieces -- [ ] ** -## Updates: - -**01/02/2023** -- New Test idea: Reconstruct missing patches -- paper to compare? Solving Jigsaw Puzzles With Eroded Boundaries -Dov Bridger, Dov Danon, Ayellet Tal; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 3526-3535 - https://openaccess.thecvf.com/content_CVPR_2020/html/Bridger_Solving_Jigsaw_Puzzles_With_Eroded_Boundaries_CVPR_2020_paper.html - - - -**31/01/2023** -- Implemented DDIM sampling-> sampling speedup -- Tested the GAT as GNN on wikiart 12x12, doesn't work well