diff --git a/.gitattributes b/.gitattributes index d6dc8b7f0e30b3129de3c30efbfdc0236c954f60..e4fedf3544038487661402a4792aed6ef70e654d 100644 --- a/.gitattributes +++ b/.gitattributes @@ -153,3 +153,7 @@ figures/textvqa_patch_tiles/split_4_3_flat.png filter=lfs diff=lfs merge=lfs -te figures/textvqa_patch_tiles/split_4_3_oblique.png filter=lfs diff=lfs merge=lfs -text figures/textvqa_patch_tiles/split_4_4_flat.png filter=lfs diff=lfs merge=lfs -text figures/textvqa_patch_tiles/split_4_4_oblique.png filter=lfs diff=lfs merge=lfs -text +figures/textvqa_patch_tiles/source_flat.png filter=lfs diff=lfs merge=lfs -text +figures/textvqa_patch_tiles/source_grid_flat.png filter=lfs diff=lfs merge=lfs -text +figures/textvqa_patch_tiles/source_grid_oblique.png filter=lfs diff=lfs merge=lfs -text +figures/textvqa_patch_tiles_src/camera_0.png filter=lfs diff=lfs merge=lfs -text diff --git a/figures/textvqa_patch_tiles/attn_0_flat.png b/figures/textvqa_patch_tiles/attn_0_flat.png index 0ed0673500ee5b65e23e852984f661d97fdd65a7..faeac26649714919c76055e8157371d5c5856004 100644 --- a/figures/textvqa_patch_tiles/attn_0_flat.png +++ b/figures/textvqa_patch_tiles/attn_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:430567dd00e687421ff7880aafddf9efaf5d6a6739d358ee7c5bebb648403959 -size 175642 +oid sha256:610b22f3435af63e7ba71ec4141b38d73fe03be0b6622834ae69080afba96d7d +size 35204 diff --git a/figures/textvqa_patch_tiles/attn_0_oblique.png b/figures/textvqa_patch_tiles/attn_0_oblique.png index a3080217a3ec1914cc4069cd5cf6ecb58c8370d7..3373b3db40763c228593d906b7d6eb21cd590f44 100644 --- a/figures/textvqa_patch_tiles/attn_0_oblique.png +++ b/figures/textvqa_patch_tiles/attn_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:299613d54387a32699dcac5d79b48b4d9bc3c5af7a5c133da337666a5dcc23ca -size 164993 +oid sha256:8d15891975247fb4203acbe7d8e645e6d74b954db206e3a3dba2fb04ec752cc1 +size 35906 diff --git a/figures/textvqa_patch_tiles/attn_1_flat.png b/figures/textvqa_patch_tiles/attn_1_flat.png index 341b6b944317c94b46a9c479502aa5533ed4907a..727acc4c6f70a75158fd317183747e13b7777dda 100644 --- a/figures/textvqa_patch_tiles/attn_1_flat.png +++ b/figures/textvqa_patch_tiles/attn_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:e180432215b63bc0d23ef9fefb06d511cd311d200d7c606b533cb4435a5890e4 -size 157918 +oid sha256:1839a5107b8d942c398adae4010c8166316b9fe4681f82a5c76f88d09553b59a +size 40360 diff --git a/figures/textvqa_patch_tiles/attn_1_oblique.png b/figures/textvqa_patch_tiles/attn_1_oblique.png index 0a756546474382f1adce267a3a703e8839937c00..77c0224e6baa817b57b42640dddcc251b993f973 100644 --- a/figures/textvqa_patch_tiles/attn_1_oblique.png +++ b/figures/textvqa_patch_tiles/attn_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:84084daa009c716fe11a1375b1324a05c221e4a1470b31a58e38b9d85fd1fdac -size 151507 +oid sha256:d736417427b8beb5cdc215d5291cc4a037a0d772bbae0f1f0d8f7f3a08f6a76d +size 40909 diff --git a/figures/textvqa_patch_tiles/attn_2_flat.png b/figures/textvqa_patch_tiles/attn_2_flat.png index 26e91410cf3b421efb7791cea7a959a228e25027..9c8d82aad2192f3a0a9ad282d6634d35f53a100c 100644 --- a/figures/textvqa_patch_tiles/attn_2_flat.png +++ b/figures/textvqa_patch_tiles/attn_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:80c30ec923d4655b2dfc4cf6b2f1872033f0a3257dd9a0fd522cee0ef2dbf4d5 -size 184006 +oid sha256:51ee95ba2003b77745f0b0c15812afd294156b63137fcd8f2a97d63f816607a6 +size 33587 diff --git a/figures/textvqa_patch_tiles/attn_2_oblique.png b/figures/textvqa_patch_tiles/attn_2_oblique.png index bf514dbba8fd9ea8eff8845ab8aad1e505c3fa29..ca54533d22c150beb43a930078dbe6c9949b5179 100644 --- a/figures/textvqa_patch_tiles/attn_2_oblique.png +++ b/figures/textvqa_patch_tiles/attn_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:74c1e2d2911c00b8ea1ab21749fd11887548b340d9bcea79ea3d073c3f8642bc -size 173921 +oid sha256:0edf56e6a90a2b15b047dbe83e2ca950ad6e082ad19ce4df46f2a693b15aced2 +size 36217 diff --git a/figures/textvqa_patch_tiles/roi_0_flat.png b/figures/textvqa_patch_tiles/roi_0_flat.png index d53e6241c89dfd1daf9d04945ec3b3d6855c4224..7f9af69faef7ca514a29430bbd172553f98bceef 100644 --- a/figures/textvqa_patch_tiles/roi_0_flat.png +++ b/figures/textvqa_patch_tiles/roi_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:de39db6cb1bdde05002e79b9c1fb613c13df27490043ed4262323a2132e876f2 -size 176632 +oid sha256:e583d9296a6fb2d3acf9e61a742c574ef55856ba11d750be53a1c2fada2b918e +size 23937 diff --git a/figures/textvqa_patch_tiles/roi_0_oblique.png b/figures/textvqa_patch_tiles/roi_0_oblique.png index bc09530b838280715cc1d762a5abe325dd9d4f07..54cd9f11461b8d6c496b2c3ec543dc54bd6f8d5a 100644 --- a/figures/textvqa_patch_tiles/roi_0_oblique.png +++ b/figures/textvqa_patch_tiles/roi_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:53d2cba89f2df68c1492feb1f936a9c5b645fc527ba9f44ee7bbf45c0d5d1f4c -size 173991 +oid sha256:496ad94fdd61dc1ee71cb6442055031f68fbaf878074920234e57ecd07576d61 +size 29108 diff --git a/figures/textvqa_patch_tiles/roi_1_flat.png b/figures/textvqa_patch_tiles/roi_1_flat.png index 6ebf1e8a275d6163ea9349bbaaddc50e7880ae30..9ef325bf51079b9c6ed670f990ca1d7f1bc30135 100644 --- a/figures/textvqa_patch_tiles/roi_1_flat.png +++ b/figures/textvqa_patch_tiles/roi_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:6b71f741df4bec09f22e5613efee045562cfec5feeebc4fac271ebfcda545b77 -size 151538 +oid sha256:b835ca7f328f73cb7804eb23817663724d8248a575dba162f463fccedaec660e +size 36499 diff --git a/figures/textvqa_patch_tiles/roi_1_oblique.png b/figures/textvqa_patch_tiles/roi_1_oblique.png index 25206641e69973ce0cd9e6d92d1e5668cbb05cfe..e4c944b1c76ccd7037c23c880d8ac789a056a806 100644 --- a/figures/textvqa_patch_tiles/roi_1_oblique.png +++ b/figures/textvqa_patch_tiles/roi_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:48f90e2e1c1b37c22a4e6e568f9b9110e4174bd58c7b03e3d09969310661f495 -size 151956 +oid sha256:3c25c840f748e11c77680bac86242a9f34acb5a465b0be9de35b20ab42b060fd +size 42598 diff --git a/figures/textvqa_patch_tiles/roi_2_flat.png b/figures/textvqa_patch_tiles/roi_2_flat.png index d07e3dcd0d0045da9f2d4ec788d84d95d4296720..6e3ca186718a23bf5ed9b84d3a2197988081be27 100644 --- a/figures/textvqa_patch_tiles/roi_2_flat.png +++ b/figures/textvqa_patch_tiles/roi_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d5cfef6723d77fe6114acfb17cda28db3dc2ada8cfe94243e431c96b510da4c9 -size 184315 +oid sha256:833e4c0e675f007dfdf2800efd8af361ed6cdb5794a057b311eb458060dd53c8 +size 26556 diff --git a/figures/textvqa_patch_tiles/roi_2_oblique.png b/figures/textvqa_patch_tiles/roi_2_oblique.png index 6df2e03e7171452935275bce77e810565f86217b..31bdb63edbea6d0917a95c30e5548b9427b0d854 100644 --- a/figures/textvqa_patch_tiles/roi_2_oblique.png +++ b/figures/textvqa_patch_tiles/roi_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:232752d3edee2493a1ad1c3ff85f8fd1919fc5e53be2f6e1da5ab36903662a56 -size 181004 +oid sha256:6d99bdb202b951f017c9ed8f7231aebc545e2fb6e88f4e7e9e77082ac83817ee +size 32405 diff --git a/figures/textvqa_patch_tiles/sink_0_flat.png b/figures/textvqa_patch_tiles/sink_0_flat.png new file mode 100644 index 0000000000000000000000000000000000000000..8c6cc80d7833271f7aef632c064bdbb3f96fc549 Binary files /dev/null and b/figures/textvqa_patch_tiles/sink_0_flat.png differ diff --git a/figures/textvqa_patch_tiles/sink_0_oblique.png b/figures/textvqa_patch_tiles/sink_0_oblique.png new file mode 100644 index 0000000000000000000000000000000000000000..a0cc23506784f05d0377b920d64c7f224a5507ee Binary files /dev/null and b/figures/textvqa_patch_tiles/sink_0_oblique.png differ diff --git a/figures/textvqa_patch_tiles/sink_1_flat.png b/figures/textvqa_patch_tiles/sink_1_flat.png new file mode 100644 index 0000000000000000000000000000000000000000..ddc0aa02d9d74b79744c868b6e107e5e729c25b0 Binary files /dev/null and b/figures/textvqa_patch_tiles/sink_1_flat.png differ diff --git a/figures/textvqa_patch_tiles/sink_1_oblique.png b/figures/textvqa_patch_tiles/sink_1_oblique.png new file mode 100644 index 0000000000000000000000000000000000000000..2d38a2b4777a30862a7f03a6995981d2fac62e81 Binary files /dev/null and b/figures/textvqa_patch_tiles/sink_1_oblique.png differ diff --git a/figures/textvqa_patch_tiles/source_flat.png b/figures/textvqa_patch_tiles/source_flat.png new file mode 100644 index 0000000000000000000000000000000000000000..70f560511b4fde023326c14e8756c5b5da50b70d --- /dev/null +++ b/figures/textvqa_patch_tiles/source_flat.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc729f3170ca23ec8d71e6051d14d4c1986c35354b15f7b694a19533b082810f +size 287235 diff --git a/figures/textvqa_patch_tiles/source_grid_flat.png b/figures/textvqa_patch_tiles/source_grid_flat.png new file mode 100644 index 0000000000000000000000000000000000000000..241b5ea8d5ac2a3db50666408d6e9f69b6259c2a --- /dev/null +++ b/figures/textvqa_patch_tiles/source_grid_flat.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb8f1fbc68d39c677189440e06201a5e2c8b0ae5c11da6268752260b5a04c7ca +size 327574 diff --git a/figures/textvqa_patch_tiles/source_grid_oblique.png b/figures/textvqa_patch_tiles/source_grid_oblique.png new file mode 100644 index 0000000000000000000000000000000000000000..e651b7e9d8200a449e8e78074225b92b0cedd350 --- /dev/null +++ b/figures/textvqa_patch_tiles/source_grid_oblique.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e889929a7cedd37b50fc5ea6d2e1ef4e31e972d30f0099b52689414f4e63994 +size 284039 diff --git a/figures/textvqa_patch_tiles/split_0_0_flat.png b/figures/textvqa_patch_tiles/split_0_0_flat.png index c8c9977083a5f5cb0fe00981a63be7304e0ba09e..373423224e170518c9b630494fb6bbf9f46a7cd0 100644 --- a/figures/textvqa_patch_tiles/split_0_0_flat.png +++ b/figures/textvqa_patch_tiles/split_0_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:46334f2b9861622d9bbecbbc4db594796a1e0c331c174f88aa9f02e9c39a74fd -size 158176 +oid sha256:2cba656a1c800a8e00272a670be61a49d9f48f1859c3d70ec4f3786f5980bf5d +size 23741 diff --git a/figures/textvqa_patch_tiles/split_0_0_oblique.png b/figures/textvqa_patch_tiles/split_0_0_oblique.png index 2ff2b2c84e47512dc5ee9f11ad135e17669208ff..cd9bc946eb359d58224c283284e4a7ec3042399f 100644 --- a/figures/textvqa_patch_tiles/split_0_0_oblique.png +++ b/figures/textvqa_patch_tiles/split_0_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:9a55642b808a501c4280e3c732e62fe8d0a69310080b2a01edd2940300c916fa -size 159279 +oid sha256:3aebbd02ca4b8415ff161f785bd75229f528c23b1a27169f803c70de18c5a4d7 +size 29279 diff --git a/figures/textvqa_patch_tiles/split_0_1_flat.png b/figures/textvqa_patch_tiles/split_0_1_flat.png index e66b1f396b4c0596501bb2262300349f993b619c..a875662aa113f0051e2ca20cf8080f6e91960abd 100644 --- a/figures/textvqa_patch_tiles/split_0_1_flat.png +++ b/figures/textvqa_patch_tiles/split_0_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c9081a06c3c41086de85899605cc64f834d4285d07f79eefff3154cf59dc27f5 -size 144482 +oid sha256:4459a674dce0cf9cd866522df22d8dd8f76e3cd3b294deb7de93f22d7f1c89cf +size 15216 diff --git a/figures/textvqa_patch_tiles/split_0_1_oblique.png b/figures/textvqa_patch_tiles/split_0_1_oblique.png index 4b35a38453d90957a41781df096d7977828c4c47..c7b6ef3fcea92a6d7359247149408a21520f5ff5 100644 --- a/figures/textvqa_patch_tiles/split_0_1_oblique.png +++ b/figures/textvqa_patch_tiles/split_0_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c0ef2d1ba694ea2ea7d7627e9b775d8b1a4cc4a546731f96a447a3f24a115961 -size 144609 +oid sha256:da1ae493ee191ebe6e042eb8f3eb08c2f3f07f52ed786d0cfcea33cc78a6faba +size 21712 diff --git a/figures/textvqa_patch_tiles/split_0_2_flat.png b/figures/textvqa_patch_tiles/split_0_2_flat.png index b278487b5799af0d62194d0451d0d956aa03ca4c..d6c32bb9ce8fb2d75fa84e3ae510a155cef7e0a0 100644 --- a/figures/textvqa_patch_tiles/split_0_2_flat.png +++ b/figures/textvqa_patch_tiles/split_0_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:eb780f6a14efa793edbfbf2b8c12c7acdf43a7fa9034da886627e9ba186b9e0c -size 171859 +oid sha256:4307fcbb791ffc6834a677c67fb92d572d76d44d07c2b5b51964ece74e032277 +size 20496 diff --git a/figures/textvqa_patch_tiles/split_0_2_oblique.png b/figures/textvqa_patch_tiles/split_0_2_oblique.png index 5a6d40a38a13dc03887601b6f9e377a603e10342..748be1e187a91d7d6d5ea1477776f4f815c656c2 100644 --- a/figures/textvqa_patch_tiles/split_0_2_oblique.png +++ b/figures/textvqa_patch_tiles/split_0_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:498250a3f4477083e3de750893d34996d3ec5bd5f077bc565840782ecc8b7a48 -size 178376 +oid sha256:74fe6d2b083e9ed03759b8d98e79c847040c38a4514dfc59aafd58423c0dd071 +size 28158 diff --git a/figures/textvqa_patch_tiles/split_0_3_flat.png b/figures/textvqa_patch_tiles/split_0_3_flat.png index 624314f81ab7b6782f4e75a9e1d5ede59bd64472..f0900c6bb94f437b2f23e3b57bff8ac0ed0035f2 100644 --- a/figures/textvqa_patch_tiles/split_0_3_flat.png +++ b/figures/textvqa_patch_tiles/split_0_3_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:9f16c817a6929ba9fc572236c7a5f7320a1b3ff41cbd37122a032109e44c5aae -size 165737 +oid sha256:89500e52fc09d88abc0f706cc29de84a45e3dcb81faa7580b92356003aca585b +size 20533 diff --git a/figures/textvqa_patch_tiles/split_0_3_oblique.png b/figures/textvqa_patch_tiles/split_0_3_oblique.png index 150a1c9551c34f65932d1bd1ad77b1089fe4d199..a02334c7f3e53d5c554c93c21eb3a657f0881ddb 100644 --- a/figures/textvqa_patch_tiles/split_0_3_oblique.png +++ b/figures/textvqa_patch_tiles/split_0_3_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:e540eacd9afb3c09b4c16bbf0df46e403d2769e26e279ca1ebf33678cb85aec6 -size 176683 +oid sha256:7b2df790a4525187aee737e1bf7b1f2321517867661a28bacfcb7e550b13bb14 +size 28004 diff --git a/figures/textvqa_patch_tiles/split_0_4_flat.png b/figures/textvqa_patch_tiles/split_0_4_flat.png index 3b47c63941a6fee4d45a62cd218dbbff85da6553..0dcbf75cf0bf50a02fffcfeadcc3a6302113ab11 100644 --- a/figures/textvqa_patch_tiles/split_0_4_flat.png +++ b/figures/textvqa_patch_tiles/split_0_4_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b69713e0cd316eed42dd5b4635187ebf7eb67e00883ea8aed93f4b7b1771ee01 -size 181543 +oid sha256:a125b18af8d6f029f7c0ac3f057f7f25bc2593bb4fd9e62f21c6652987e645f9 +size 14106 diff --git a/figures/textvqa_patch_tiles/split_0_4_oblique.png b/figures/textvqa_patch_tiles/split_0_4_oblique.png index 2257df68461cdb78273f082b878efd4c4684791c..8ff803e222b5612e122110038512437fdf37e46a 100644 --- a/figures/textvqa_patch_tiles/split_0_4_oblique.png +++ b/figures/textvqa_patch_tiles/split_0_4_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:3b97c0daab9e603db88b847cec4c4f1169ad16fbcca6540ea4c01c2fac55c2ed -size 189038 +oid sha256:05790ea1fd7a1bce1dc22e310c55bbd5f678ec39da6ccce39f7e9f1e3e170c1a +size 22045 diff --git a/figures/textvqa_patch_tiles/split_1_0_flat.png b/figures/textvqa_patch_tiles/split_1_0_flat.png index ead51bf94f9bda3e8a441e384f6d0397a389e242..5ba149ecd4b834bf6df18bea718a5050b30bfdec 100644 --- a/figures/textvqa_patch_tiles/split_1_0_flat.png +++ b/figures/textvqa_patch_tiles/split_1_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:31b96d2fd5003bf1fb2e54b7ea03a39df20a044df9fff7ababafd7df6a83bc9a -size 162519 +oid sha256:1b073e07cd990e05c4cfbef0ba3ddfeb84052f816ec5325696b8697056f0b1b3 +size 26601 diff --git a/figures/textvqa_patch_tiles/split_1_0_oblique.png b/figures/textvqa_patch_tiles/split_1_0_oblique.png index c0d477c2f9aafb44dde76c0520a15aaa662fc69e..f51db306f0828f45a509a1f65d7466a06afe7432 100644 --- a/figures/textvqa_patch_tiles/split_1_0_oblique.png +++ b/figures/textvqa_patch_tiles/split_1_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:a2d09a52f0c5ed098eab5c17197b6cc50d3879ba0037e0c8d27821a84e6dda61 -size 163880 +oid sha256:b6e1395b0febffd7f14ac4aa177f42032e26e376c5eeae68f939dcf7895e50e0 +size 27056 diff --git a/figures/textvqa_patch_tiles/split_1_1_flat.png b/figures/textvqa_patch_tiles/split_1_1_flat.png index 87075136c1339edc855c8ad329365f5c8e97fdb5..909c5ed8037f3491d6a2a102663d66763af4c9dd 100644 --- a/figures/textvqa_patch_tiles/split_1_1_flat.png +++ b/figures/textvqa_patch_tiles/split_1_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:3388c3237620dfdce0e34ca4dddff2e797d1b8ed6e0c15024e255606cd5ba7bf -size 170785 +oid sha256:44dbda265f93e0428c6808fa52ade1361960644d5f7af80d97b739adba47400d +size 40875 diff --git a/figures/textvqa_patch_tiles/split_1_1_oblique.png b/figures/textvqa_patch_tiles/split_1_1_oblique.png index 13d5ef54486c9afcbb4123e5d595f4b9e708966e..90ce599139994d3cd32366de04e0619998142f8d 100644 --- a/figures/textvqa_patch_tiles/split_1_1_oblique.png +++ b/figures/textvqa_patch_tiles/split_1_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:ba19fc2ef4f3a366ce695021b9dcde5b7a357a0375c23280edc74996c30905ae -size 172982 +oid sha256:3b108b9cec6fb4be3e5e25fd7d8eeffa5809c1182a7b4f512bbc261171134c3a +size 42234 diff --git a/figures/textvqa_patch_tiles/split_1_2_flat.png b/figures/textvqa_patch_tiles/split_1_2_flat.png index a749b0749558afc761fc2ca46389f45d568d4b7c..8de186ca4c14895bec2d4599f149673bf6da75f7 100644 --- a/figures/textvqa_patch_tiles/split_1_2_flat.png +++ b/figures/textvqa_patch_tiles/split_1_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:644f7189fa0f501ed27b66677a2f81050f007cd17e29dae425653daca065f67a -size 208695 +oid sha256:2898baa3bd67c0fc0d2f81b8ffedcde63cdb0fadb6fcbd00c589a38fb21906b8 +size 29007 diff --git a/figures/textvqa_patch_tiles/split_1_2_oblique.png b/figures/textvqa_patch_tiles/split_1_2_oblique.png index e3e8632beeaba919ce97896528a8a065dd29d4cf..8317cb39765f9bb7e0a87b0a5fd1b5f258b0f6ac 100644 --- a/figures/textvqa_patch_tiles/split_1_2_oblique.png +++ b/figures/textvqa_patch_tiles/split_1_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c4f16fbec42183b793d1cbf3fb0cca9d9fc9333dc17af52808e291802aa9aafa -size 200300 +oid sha256:f6bab5449b06d6099bef6e92c3bc83cd0d266fdfc4e22323b69d723a80371be5 +size 29405 diff --git a/figures/textvqa_patch_tiles/split_1_3_flat.png b/figures/textvqa_patch_tiles/split_1_3_flat.png index fed0d9853a7f690e43895757383d61e4e5baae16..cf235b56082f8858ac200efd39b04529ecae3b56 100644 --- a/figures/textvqa_patch_tiles/split_1_3_flat.png +++ b/figures/textvqa_patch_tiles/split_1_3_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:9ff8b83241a9603c11309ca58d7d44454731193c75f55a1f345e7a92497d188b -size 210556 +oid sha256:8cd658e2e8036c7e1a555d7913247c1925c2a02e1638a2e02dc3803693299eb2 +size 28206 diff --git a/figures/textvqa_patch_tiles/split_1_3_oblique.png b/figures/textvqa_patch_tiles/split_1_3_oblique.png index 4fe47f2e334eba9d4815b79c7b0220816ba662a2..35befd94ace8eabbf9a2c7c166e3aaaaa96404e2 100644 --- a/figures/textvqa_patch_tiles/split_1_3_oblique.png +++ b/figures/textvqa_patch_tiles/split_1_3_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:93cee4e7faaab210b8e30e0e0b6ef80bd266b01780f80645991a87d9ab8f816e -size 196702 +oid sha256:6f06ca6d8906777c5796efd3728e8ed2a5945aad77b785fa484a36ea61966d74 +size 28310 diff --git a/figures/textvqa_patch_tiles/split_1_4_flat.png b/figures/textvqa_patch_tiles/split_1_4_flat.png index cff12f3bd5d97f6f75379734926f51f4c775346a..878e90acdde528f2935a6260595ec209572e2103 100644 --- a/figures/textvqa_patch_tiles/split_1_4_flat.png +++ b/figures/textvqa_patch_tiles/split_1_4_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:481a2cd644ead651522959c717c558c8791411ce8bf0277d88378c84d32fb07d -size 197397 +oid sha256:996755f7e65d67d58f8b7ebb9f62b5bfbba4426a79b6942464997b467ee2d548 +size 27356 diff --git a/figures/textvqa_patch_tiles/split_1_4_oblique.png b/figures/textvqa_patch_tiles/split_1_4_oblique.png index c4f81dabb1818cea9973a16908b7f5b64c259cfa..b807e477a596da7452bd70efc8d9cfe71dadfd43 100644 --- a/figures/textvqa_patch_tiles/split_1_4_oblique.png +++ b/figures/textvqa_patch_tiles/split_1_4_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b6ab18b1af6952e5d357a421877b2c383392e759194669788c719b3c1a39c93c -size 187068 +oid sha256:51f434b17c6658fde979aef9f031c8f2ebba3879be5ded869a2bb747f23069fd +size 30514 diff --git a/figures/textvqa_patch_tiles/split_2_0_flat.png b/figures/textvqa_patch_tiles/split_2_0_flat.png index 0e20f2e5bf1f01862a05ae3085166cd2b1573618..e6630f688943a297bdf2718958b435174caf0434 100644 --- a/figures/textvqa_patch_tiles/split_2_0_flat.png +++ b/figures/textvqa_patch_tiles/split_2_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:ae54856b128a02e591a3944da859997127a2a9f0b1e961a78aaea5c2b727372c -size 174831 +oid sha256:f41fa77a2a3d800e0c135aaa83b6186baf0c72c8afe0989572877038bcbc3ccb +size 27610 diff --git a/figures/textvqa_patch_tiles/split_2_0_oblique.png b/figures/textvqa_patch_tiles/split_2_0_oblique.png index ae58aef753facf6550c2c5ff01ad1814b69e595e..e7e20c9053c044c939d45385721292a76de6f65a 100644 --- a/figures/textvqa_patch_tiles/split_2_0_oblique.png +++ b/figures/textvqa_patch_tiles/split_2_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:e832465ee6493a1cc7409f68bd4e8793bcdbb407a5f53ab19c0ee83582c5dffe -size 171160 +oid sha256:4635ce86c4b8473bef528cfb6f2e2d2203b9638e89ca59640e87e950b2da1b2c +size 32494 diff --git a/figures/textvqa_patch_tiles/split_2_1_flat.png b/figures/textvqa_patch_tiles/split_2_1_flat.png index 4d74cad33252d84d80ea76c1bfd90778bbd88976..c33675037e20e75c3d04003f02c760eda8f966f1 100644 --- a/figures/textvqa_patch_tiles/split_2_1_flat.png +++ b/figures/textvqa_patch_tiles/split_2_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:cd82bc4cd86b971ec9c5423c664227de6124eb677b9adf810806c5923fd85705 -size 199174 +oid sha256:c8ca89be03459b63c8a968b12c08586279d91364c14833bfc882e62688d0236d +size 25519 diff --git a/figures/textvqa_patch_tiles/split_2_1_oblique.png b/figures/textvqa_patch_tiles/split_2_1_oblique.png index 3227736d9baf73a3e7d3b0b190828dc994c054be..5f52813c7232eada015e4f52cbbc951b896cb333 100644 --- a/figures/textvqa_patch_tiles/split_2_1_oblique.png +++ b/figures/textvqa_patch_tiles/split_2_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:078e0b38e23b2555cac0165c2e085994d96ef720db77a9f825fc8b504466f7b1 -size 187483 +oid sha256:e1b0923947a5c5e672de10c72eaa3540e9c0db0228a52400000375c9db608a2e +size 28521 diff --git a/figures/textvqa_patch_tiles/split_2_2_flat.png b/figures/textvqa_patch_tiles/split_2_2_flat.png index 49176d157714711ac26074036b58b34dc94f564e..334e7862df17a5925c9a7ed76ffc260bfdf5ca95 100644 --- a/figures/textvqa_patch_tiles/split_2_2_flat.png +++ b/figures/textvqa_patch_tiles/split_2_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b5b73d59f946173dc0f64c898ce8324939563e1ab672181b77784300047e6135 -size 171992 +oid sha256:b63669dfd1b9c117dce4f08e2b6eb6c3fb0cbb441af8a1b99504c4499d09f633 +size 9590 diff --git a/figures/textvqa_patch_tiles/split_2_2_oblique.png b/figures/textvqa_patch_tiles/split_2_2_oblique.png index 67819e0bfd63b2481535ca4389d8388e70cab518..f22de78b6aca013ff4ab9fb47e357bd8a7b0bed9 100644 --- a/figures/textvqa_patch_tiles/split_2_2_oblique.png +++ b/figures/textvqa_patch_tiles/split_2_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:fc99b6e4e448f14fb7e4dcc13eed8f308aa94108dce93081e3ace88c038991f9 -size 163420 +oid sha256:2a63c36281a96fdcbe104372df058764f9486c9de479b1f74313bb3c3331d31f +size 15157 diff --git a/figures/textvqa_patch_tiles/split_2_3_flat.png b/figures/textvqa_patch_tiles/split_2_3_flat.png index 71ebaf91c00f72e9a4a8343ee5c9b5f29b6f33cd..3831c4763b4a1794f018007f0806d7f6bb9c1b78 100644 --- a/figures/textvqa_patch_tiles/split_2_3_flat.png +++ b/figures/textvqa_patch_tiles/split_2_3_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b12fd2fb75a5f57eb976c8be7b8d898ad4dbe09e2f1c812ef9af029fb124c445 -size 210243 +oid sha256:722b77fa12ec4e1676f3d4789667fe8b6cb49c7aea5460284de068cb74b8fd40 +size 11015 diff --git a/figures/textvqa_patch_tiles/split_2_3_oblique.png b/figures/textvqa_patch_tiles/split_2_3_oblique.png index 43d90e8fdafebce730dc557a75527d24923504e1..21ed146bf4036123e64b9e8104f09480b24a9a80 100644 --- a/figures/textvqa_patch_tiles/split_2_3_oblique.png +++ b/figures/textvqa_patch_tiles/split_2_3_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:ac052fe9b120594997d8de8469ea3c7ce6ad6de4377cd6200bcdc960f4673693 -size 195431 +oid sha256:185c8254acec9b8355e12b866e2c4df0f78e06b4842f11886fdae001e91c978b +size 17102 diff --git a/figures/textvqa_patch_tiles/split_2_4_flat.png b/figures/textvqa_patch_tiles/split_2_4_flat.png index e50ef333f0d98ea82ce7f87fa4938664b0aa7a0d..dcc6d1d7c08fa8bbbfa3ca93030d98b7b3655753 100644 --- a/figures/textvqa_patch_tiles/split_2_4_flat.png +++ b/figures/textvqa_patch_tiles/split_2_4_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:0d62803547ca5165c343ecbe20e005f197ee73dada4cb4ef9c016f4767cca1d6 -size 198840 +oid sha256:e8edea52bc7eebccf33ee899fd141aeefc9c1379c204b4a49e88754107954dcd +size 30041 diff --git a/figures/textvqa_patch_tiles/split_2_4_oblique.png b/figures/textvqa_patch_tiles/split_2_4_oblique.png index cca356dee2c89798a0cc812f06499ac58d651155..2de34703b416b62559cab03e9fab4bcd8e35aa1b 100644 --- a/figures/textvqa_patch_tiles/split_2_4_oblique.png +++ b/figures/textvqa_patch_tiles/split_2_4_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:e24c9784286e2e784fb1cc6768b3794a9973e68a165057f824ea8feb14a9cb58 -size 195002 +oid sha256:686e1419ba0dc815b1b902e8792ba0cf95c264e963dde447680032d609159ec8 +size 34254 diff --git a/figures/textvqa_patch_tiles/split_3_0_flat.png b/figures/textvqa_patch_tiles/split_3_0_flat.png index 490b36193a1a9cb1c1bf74dd8dbd9427ee90e6d5..a88739c692434e9fcbe038fd9220eb899689d6d8 100644 --- a/figures/textvqa_patch_tiles/split_3_0_flat.png +++ b/figures/textvqa_patch_tiles/split_3_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d54cb7aa9b43d71ff6b8995bbfc19dbb281d769f9bbf6c6259c5aaeadb564db6 -size 146056 +oid sha256:1f1f0bb365e8cc94b600b8a0beda371bae2a58e2de29a465511e7fac9a93e671 +size 25823 diff --git a/figures/textvqa_patch_tiles/split_3_0_oblique.png b/figures/textvqa_patch_tiles/split_3_0_oblique.png index b64248a5dcae2053502a6428e6d0fbcaa6072695..38d1e4374c6061c8fbe2744b3996c1967acb2ff9 100644 --- a/figures/textvqa_patch_tiles/split_3_0_oblique.png +++ b/figures/textvqa_patch_tiles/split_3_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:4e3e6c7c3c976e2599ea0e01b8b84df8790416a37330c0ab5fb51b70aff8e3e5 -size 151710 +oid sha256:ae211ca1b2bc8ad40117f88d96a28a6eba80f771ece68db82afab34124e8d763 +size 29984 diff --git a/figures/textvqa_patch_tiles/split_3_1_flat.png b/figures/textvqa_patch_tiles/split_3_1_flat.png index 6ae3730a37310223181810cdbe38cd7eec04727b..bd8d7c73bcf1ce2c84947d5ba4088271db3b71eb 100644 --- a/figures/textvqa_patch_tiles/split_3_1_flat.png +++ b/figures/textvqa_patch_tiles/split_3_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:17e1b3d41572aa297598bcd1baa496b55670b651f8af98c59dc5bb2905bc0ed8 -size 160056 +oid sha256:a2a64119e5ea6f55623a8072bec16c8ebfe82164ca8b6a97d110cefe99974bd1 +size 35827 diff --git a/figures/textvqa_patch_tiles/split_3_1_oblique.png b/figures/textvqa_patch_tiles/split_3_1_oblique.png index c80eedc961f49a69f223dc98d504392ee6fd1b81..6a9f6dfaec47255868b9f5da2ce4e3687f426f6f 100644 --- a/figures/textvqa_patch_tiles/split_3_1_oblique.png +++ b/figures/textvqa_patch_tiles/split_3_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:2ae8e9be9e0df70c746bc506db635071315e9f027710e670c7ef32cfbe4eaff4 -size 161943 +oid sha256:6debbe581bf6e83f812078ce3000540baa3a33566f7549afa2eaecd7bcd393f5 +size 36085 diff --git a/figures/textvqa_patch_tiles/split_3_2_flat.png b/figures/textvqa_patch_tiles/split_3_2_flat.png index ebb70911ecb7babe138e8cb291a9b42b02865295..050c1b996ad07cead32d0a260cd27b6678a4dc36 100644 --- a/figures/textvqa_patch_tiles/split_3_2_flat.png +++ b/figures/textvqa_patch_tiles/split_3_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:a5571c805069246ef0eab479df57b4b2c828330dc4e8393dbd3c2b894be6bc4b -size 150427 +oid sha256:61eb66aafc3543520a2c4fd8c89ac445716deadcbd56984cc136f8ee9ae1a8a7 +size 20926 diff --git a/figures/textvqa_patch_tiles/split_3_2_oblique.png b/figures/textvqa_patch_tiles/split_3_2_oblique.png index 92d83ea5481ae6084214ccd9520a5dae5f95246a..9acf64a79a8d24c5277fcc034d2b9f21d4652797 100644 --- a/figures/textvqa_patch_tiles/split_3_2_oblique.png +++ b/figures/textvqa_patch_tiles/split_3_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:0757b25d658ce46cbf87b8582c9b2d59f4845b62682c206beb55c594902ac541 -size 158112 +oid sha256:b58ab96809d014216ab19ba4390edf8ea20fe87eb816514b2690be30b54671a4 +size 28259 diff --git a/figures/textvqa_patch_tiles/split_3_3_flat.png b/figures/textvqa_patch_tiles/split_3_3_flat.png index fd0635cdd0b997822beb264e7784041e52c3737f..de0570bd444c8ad8028852a2232d99a6b46ae891 100644 --- a/figures/textvqa_patch_tiles/split_3_3_flat.png +++ b/figures/textvqa_patch_tiles/split_3_3_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:bc9f470d2c0f0a03d52cf40935f42973cdc335c8ed70a7c010de95319ac1dacd -size 139207 +oid sha256:794197ea322c6cfe8c6413f66f67606ae8b6cdbcd9e52d7e063bd9407abd9ba0 +size 11628 diff --git a/figures/textvqa_patch_tiles/split_3_3_oblique.png b/figures/textvqa_patch_tiles/split_3_3_oblique.png index ee99b8667427e72e23a20766413ea9bc91d1fcc1..8ee71986c016494872596f6cd363bd27036369d4 100644 --- a/figures/textvqa_patch_tiles/split_3_3_oblique.png +++ b/figures/textvqa_patch_tiles/split_3_3_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d5ce604e1a3a3e78de39417b62ecb12f832470708736c587dcf5ef6678180c43 -size 151190 +oid sha256:7d0bb5fc2482b1c00094c66939386d488d31968d50183efff2d518859b4a88fa +size 17618 diff --git a/figures/textvqa_patch_tiles/split_3_4_flat.png b/figures/textvqa_patch_tiles/split_3_4_flat.png index fd5b56bf85a09d0f1c5de1a8f061b29509fdb422..c6bb4fab9bf1b2ac770a729af52cc9296f74afdc 100644 --- a/figures/textvqa_patch_tiles/split_3_4_flat.png +++ b/figures/textvqa_patch_tiles/split_3_4_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b5e740642116aa1efb5a56a5e4c50955926296975d6879581b40d5f815714af4 -size 168667 +oid sha256:044d0f329bb6ae9ddf8e766e1796d80a116a72b87c72b2b48d60b63101d51433 +size 26879 diff --git a/figures/textvqa_patch_tiles/split_3_4_oblique.png b/figures/textvqa_patch_tiles/split_3_4_oblique.png index ef4dc1fb170745f2671b2ee5de74212d84822db5..889da6848eb0708a6f02244c5fc2368b5ddbebef 100644 --- a/figures/textvqa_patch_tiles/split_3_4_oblique.png +++ b/figures/textvqa_patch_tiles/split_3_4_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d22a0bfee925280bd5875c63f3c277a895837f0eb9d5cef7f767efc8286e4ce2 -size 152415 +oid sha256:0b6e9413547228e364c8acc72c6fc3371da2ecbf62a36abca596156e4b51532f +size 29290 diff --git a/figures/textvqa_patch_tiles/split_4_0_flat.png b/figures/textvqa_patch_tiles/split_4_0_flat.png index 01ce1b0f4190f0def1f3782aeea6b9cad1f34f9f..76006ef77c2d2a674900365d4eb4680ac3371a98 100644 --- a/figures/textvqa_patch_tiles/split_4_0_flat.png +++ b/figures/textvqa_patch_tiles/split_4_0_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c82018f5915d30f73c6b9833ea779d0766c1d64fab6d24872c93af81e15a0226 -size 126397 +oid sha256:d898cae1f1332eec11088a73674283e7bdd1ee7841d813ffd39980b5f62434cd +size 18462 diff --git a/figures/textvqa_patch_tiles/split_4_0_oblique.png b/figures/textvqa_patch_tiles/split_4_0_oblique.png index 58eba926ff94df44d4b4660380b40cd23ae93332..0ee8078c39bd8e502b1ba22bdacebe15785583c3 100644 --- a/figures/textvqa_patch_tiles/split_4_0_oblique.png +++ b/figures/textvqa_patch_tiles/split_4_0_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:ee13efe61755d2d7e7c79bd33921599b7e5cbfd2328981d82c44ff42e0861ef0 -size 131020 +oid sha256:274129747607581b44bbe9b5eec21fdbd196766a721f63282cef7b59c8f85e30 +size 21021 diff --git a/figures/textvqa_patch_tiles/split_4_1_flat.png b/figures/textvqa_patch_tiles/split_4_1_flat.png index 2bba1e1e8bcb96420deb74a9e6421418e93a1f94..7db62a2a5dcd71a7079354922ba0627ab9918440 100644 --- a/figures/textvqa_patch_tiles/split_4_1_flat.png +++ b/figures/textvqa_patch_tiles/split_4_1_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:273220e319f358bc33b9d8d8bf11180c68f6a772795706aab1a617b550820284 -size 161700 +oid sha256:ecc4aa39b3147952074a5133581fbf4e83437ac47ca813b594a3ac614f9ae968 +size 10737 diff --git a/figures/textvqa_patch_tiles/split_4_1_oblique.png b/figures/textvqa_patch_tiles/split_4_1_oblique.png index 9a7a454cbceb0e91c273e07d0b14004d16e714a6..f391bc529b0027f642436a321b3051374c70a0b3 100644 --- a/figures/textvqa_patch_tiles/split_4_1_oblique.png +++ b/figures/textvqa_patch_tiles/split_4_1_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d2ce56f956ea537d4c21ae717712371997151e98b4e89ac296204e3e5a5f8692 -size 157211 +oid sha256:ed9118793de084ef0f7a57f1bc163436f53ea23345fd052c622cefb74fe68ecb +size 17420 diff --git a/figures/textvqa_patch_tiles/split_4_2_flat.png b/figures/textvqa_patch_tiles/split_4_2_flat.png index 1e6a8f61a1ee3b28956043a3c42c957acb0fc35f..62e9165174e3d7df5847180d098e661154d00afd 100644 --- a/figures/textvqa_patch_tiles/split_4_2_flat.png +++ b/figures/textvqa_patch_tiles/split_4_2_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:7bf32407da90f93d3f864cd51d00321feb859428a4683c79fe165fd279d13992 -size 163181 +oid sha256:c63532abb3ab5d297833d4e098b1aae1886de4502217863a31fdf1b04d2e1915 +size 16628 diff --git a/figures/textvqa_patch_tiles/split_4_2_oblique.png b/figures/textvqa_patch_tiles/split_4_2_oblique.png index 1474f0389a5b466f777d748300c05f8d67bc7c3c..3542b56043ffe8b6ea5697663f08787dd33d88cf 100644 --- a/figures/textvqa_patch_tiles/split_4_2_oblique.png +++ b/figures/textvqa_patch_tiles/split_4_2_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:940fd8982d9ed75281b87d049929e6930ebf492c0dba4400b62bd05f07b4da4e -size 156823 +oid sha256:7673c6a811474f309883282a56f14356bb2917574aa13f6a0fbe5d51be2125a5 +size 20944 diff --git a/figures/textvqa_patch_tiles/split_4_3_flat.png b/figures/textvqa_patch_tiles/split_4_3_flat.png index 57c24bf65ad6c8612c1fd59a5182c3c68565f65c..fd2d88c8c6f87a313eb86de28ea38692427e5fe7 100644 --- a/figures/textvqa_patch_tiles/split_4_3_flat.png +++ b/figures/textvqa_patch_tiles/split_4_3_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:3784c814b9d31295926fd13db88aa49bc2165c82005774b683cf8c187da92c3a -size 126409 +oid sha256:8f477fcb2cdfe2c5f1cd207e2609eb312bb3798949c74f8140c734f325f395b3 +size 22455 diff --git a/figures/textvqa_patch_tiles/split_4_3_oblique.png b/figures/textvqa_patch_tiles/split_4_3_oblique.png index 1cb56a5353d92622cc7c6f8c8787a5b0dec22a6a..113a7f116550a0a3e0e045bc209805cddcb3c05b 100644 --- a/figures/textvqa_patch_tiles/split_4_3_oblique.png +++ b/figures/textvqa_patch_tiles/split_4_3_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:4259f8aba0aa3248db2e1e6bc5a2cee23f9bed293dc7413efddf7bc20fe0e0fc -size 130306 +oid sha256:eeb247c9afbf3400045e085288a647d6f2759b64d65bf1dc3bf69867abb1e56e +size 25348 diff --git a/figures/textvqa_patch_tiles/split_4_4_flat.png b/figures/textvqa_patch_tiles/split_4_4_flat.png index 75c3303723bd576d35587810baee131234d5bffd..60d71ee1b42c861b45c4c5250b6a46cb051d79be 100644 --- a/figures/textvqa_patch_tiles/split_4_4_flat.png +++ b/figures/textvqa_patch_tiles/split_4_4_flat.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:a0480803c63356fcb33ff3b56beb7b65d9e8ebb9a7cfbbff1c98dfe0ed0c6ce3 -size 121071 +oid sha256:94bfe4c95b5183341f0c7aeb19239379e159b26e9fc1e43069a2b1ec85dcd901 +size 22410 diff --git a/figures/textvqa_patch_tiles/split_4_4_oblique.png b/figures/textvqa_patch_tiles/split_4_4_oblique.png index 5b22fee1249c572d341776b6f75a77fdf5060b63..16136391d0528dac8938226a65dc4345d6588c2d 100644 --- a/figures/textvqa_patch_tiles/split_4_4_oblique.png +++ b/figures/textvqa_patch_tiles/split_4_4_oblique.png @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:f3642234a3d6cd0d647709920f2a56ba9d624b7a66eea8a788c934b498763199 -size 125858 +oid sha256:ff04aedd7f5c27b2cc76056d8e70c64d4c0e85ac73d68b8a8bea3d20f11e2e4f +size 27859 diff --git a/figures/textvqa_patch_tiles_src/camera_0.png b/figures/textvqa_patch_tiles_src/camera_0.png new file mode 100644 index 0000000000000000000000000000000000000000..db9e0652b1c5a17c01046c1f0a61724cc3c464dd --- /dev/null +++ b/figures/textvqa_patch_tiles_src/camera_0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8d3148438f36a71b5577546df72094ca3f1dc4c09b745e02a1352fd526a5212 +size 442435 diff --git a/scripts/gen_textvqa_patch_figures.py b/scripts/gen_textvqa_patch_figures.py index 9bad7b72adcd9a9a21a4c7f895653e81b8fa3ad2..bfb6d1c8b5526012396eda341a8f8a9c2fa3072e 100644 --- a/scripts/gen_textvqa_patch_figures.py +++ b/scripts/gen_textvqa_patch_figures.py @@ -19,7 +19,7 @@ from PIL import Image, ImageDraw, ImageFilter # ------------------------------------------------------------------------- # Source + outputs # ------------------------------------------------------------------------- -TEXTVQA_IMG = "/mnt/bn/leonworkspace/terry/data/textvqa/train_images/0000599864fd15b3.jpg" +TEXTVQA_IMG = "/opt/tiger/thothvl_pretrain/doc/figures/textvqa_patch_tiles_src/camera_0.png" NAS_OUT = "/mnt/bn/leonworkspace/terry/ce-task/figures/textvqa_patch_tiles" LOCAL_OUT = "/opt/tiger/thothvl_pretrain/doc/figures/textvqa_patch_tiles" @@ -27,29 +27,34 @@ LOCAL_OUT = "/opt/tiger/thothvl_pretrain/doc/figures/textvqa_patch_tiles" # Config # ------------------------------------------------------------------------- GRID = 5 # 5x5 patch grid -PATCH_PX = 360 # source pixels per patch (patch is square) +PATCH_PX = 180 # source pixels per patch (patch is square) # Oblique tile canvas -TILE_W, TILE_H = 480, 400 -SHADOW_BLUR = 14 -SHADOW_OFFSET = (12, 20) +TILE_W, TILE_H = 240, 200 +SHADOW_BLUR = 7 +SHADOW_OFFSET = (6, 10) SHADOW_OPACITY = 110 SHADOW_COLOR = (60, 65, 75) # Flat 2D tile canvas (no perspective) -FLAT_W, FLAT_H = 380, 380 -FLAT_PADDING = 10 # pixels of canvas margin around the patch - -# ROI / attention -# (row, col) of M query-relevant patches in the 5x5 grid. Picked manually -# for the chosen TextVQA image so the paper figure shows a coherent region -# (e.g. a sign / receipt cluster); change if swapping the source image. -ROI_PATCHES = [(2, 1), (2, 2), (2, 3), (3, 2), (1, 2)] +FLAT_W, FLAT_H = 190, 190 +FLAT_PADDING = 5 # pixels of canvas margin around the patch + +# ROI / sink / attention +# (row, col) of query-relevant patches and sink-token patches in the 5x5 +# grid. Indices are mapped from the 24x24 attention grid in the +# pseudo-label visualization for /textvqa/0_pseudo_label.png: +# FG region (yellow) → grid x[4..9], y[5..9] → patches (1, 0..2) +# Sink token (gray Ignore) → grid x[4], y[15] → patches (3, 0..1) +ROI_PATCHES = [(1, 0), (1, 1), (1, 2)] +SINK_PATCHES = [(3, 0), (3, 1)] M = len(ROI_PATCHES) +N_SINK = len(SINK_PATCHES) -ROI_BORDER_COLOR = (235, 60, 50, 255) # crimson -ROI_BORDER_WIDTH = 14 # px on the source patch -HEATMAP_ALPHA = 0.55 # 0..1 overlay strength +ROI_BORDER_COLOR = (235, 60, 50, 255) # crimson — query-relevant ROI +SINK_BORDER_COLOR = (90, 110, 140, 255) # slate-blue — sink token +BORDER_WIDTH = 7 # px on the source patch +HEATMAP_ALPHA = 0.55 # 0..1 attention overlay strength # ------------------------------------------------------------------------- # Geometry @@ -120,7 +125,7 @@ def synth_heatmap(patch: Image.Image, peak_xy: tuple[float, float], def add_border(patch: Image.Image, color=ROI_BORDER_COLOR, - width: int = ROI_BORDER_WIDTH) -> Image.Image: + width: int = BORDER_WIDTH) -> Image.Image: """Draw a thick coloured rectangle just inside the patch edges.""" out = patch.convert("RGBA").copy() d = ImageDraw.Draw(out) @@ -170,17 +175,28 @@ def render_flat(patch_img: Image.Image) -> Image.Image: # Save helpers # ------------------------------------------------------------------------- def save_pair(im: Image.Image, name: str): - """Save PNG (with alpha) and PDF (flattened on white) at 300 DPI.""" - pdf_im = Image.new("RGB", im.size, (255, 255, 255)) - pdf_im.paste(im, mask=im.split()[3] if im.mode == "RGBA" else None) + """Save PNG with alpha (modest resolution for paper figures).""" for d in (NAS_OUT, LOCAL_OUT): - im.save(os.path.join(d, f"{name}.png"), dpi=(300, 300)) - pdf_im.save(os.path.join(d, f"{name}.pdf"), "PDF", resolution=300.0) + im.save(os.path.join(d, f"{name}.png"), dpi=(150, 150)) # ------------------------------------------------------------------------- # Main # ------------------------------------------------------------------------- +def draw_grid(img: Image.Image, n: int = GRID, + color=(255, 255, 255, 230), width: int = 4) -> Image.Image: + """Draw n×n grid lines on top of a square source image.""" + out = img.convert("RGBA").copy() + d = ImageDraw.Draw(out) + w, h = out.size + for k in range(1, n): + x = round(k * w / n) + y = round(k * h / n) + d.line([(x, 0), (x, h)], fill=color, width=width) + d.line([(0, y), (w, y)], fill=color, width=width) + return out + + def render_all(): os.makedirs(NAS_OUT, exist_ok=True) os.makedirs(LOCAL_OUT, exist_ok=True) @@ -189,6 +205,40 @@ def render_all(): sq = load_square_source(TEXTVQA_IMG, src_size) print(f"[src] {TEXTVQA_IMG} -> {src_size}x{src_size} square") + # 0) Full source image with 5×5 grid overlay — flat + oblique at a + # larger canvas than the per-patch tiles so grid lines stay readable. + full_grid = draw_grid(sq) + full_flat = full_grid.resize((600, 600), Image.LANCZOS) + save_pair(full_flat, "source_grid_flat") + + big_w, big_h = 720, 600 + dst = [ + (big_w * 0.235, big_h * 0.080), + (big_w * 0.765, big_h * 0.080), + (big_w * 0.945, big_h * 0.840), + (big_w * 0.055, big_h * 0.840), + ] + src = [(0, 0), (full_grid.width, 0), + (full_grid.width, full_grid.height), (0, full_grid.height)] + coeffs = perspective_coeffs(src, dst) + warped = full_grid.convert("RGBA").transform( + (big_w, big_h), Image.PERSPECTIVE, coeffs, resample=Image.BICUBIC) + bg = Image.new("RGBA", (big_w, big_h), (0, 0, 0, 0)) + sm = Image.new("L", (big_w, big_h), 0) + ImageDraw.Draw(sm).polygon( + [(x + SHADOW_OFFSET[0] * 2, y + SHADOW_OFFSET[1] * 2) for x, y in dst], + fill=SHADOW_OPACITY) + sm = sm.filter(ImageFilter.GaussianBlur(SHADOW_BLUR * 2)) + sl = np.zeros((big_h, big_w, 4), dtype=np.uint8) + sl[:, :, 0:3] = SHADOW_COLOR + sl[:, :, 3] = np.asarray(sm, dtype=np.uint8) + bg = Image.alpha_composite(bg, Image.fromarray(sl, "RGBA")) + bg = Image.alpha_composite(bg, warped) + save_pair(bg, "source_grid_oblique") + # Also save the bare source (no grid) at the same large canvas. + save_pair(sq.resize((600, 600), Image.LANCZOS), "source_flat") + print(f" source — flat + oblique (with and without grid)") + # 1) Full 5x5 split — every patch as oblique + flat. for i in range(GRID): for j in range(GRID): @@ -204,7 +254,7 @@ def render_all(): patch = sq.crop((x0, y0, x0 + PATCH_PX, y0 + PATCH_PX)) # ROI = patch with crimson border (the highlight) - roi = add_border(patch) + roi = add_border(patch, color=ROI_BORDER_COLOR) save_pair(render_oblique(roi), f"roi_{k}_oblique") save_pair(render_flat(roi), f"roi_{k}_flat") @@ -216,7 +266,16 @@ def render_all(): save_pair(render_flat(att), f"attn_{k}_flat") print(f" ROI {k} = grid({i},{j}) — roi + attn (oblique + flat)") - print(f"[done] wrote split + ROI + attn → {NAS_OUT}") + # 4) Sink-token patches highlighted with a distinct slate-blue border. + for k, (i, j) in enumerate(SINK_PATCHES): + x0, y0 = j * PATCH_PX, i * PATCH_PX + patch = sq.crop((x0, y0, x0 + PATCH_PX, y0 + PATCH_PX)) + sink = add_border(patch, color=SINK_BORDER_COLOR) + save_pair(render_oblique(sink), f"sink_{k}_oblique") + save_pair(render_flat(sink), f"sink_{k}_flat") + print(f" SINK {k} = grid({i},{j}) — sink (oblique + flat)") + + print(f"[done] wrote split + ROI + attn + sink → {NAS_OUT}") if __name__ == "__main__":