Updating ArtCNN

The new shaders are under the jax branch for now, but they'll go to the main branch soon. Also removing linear-light upsampling section. All shaders under test were trained with images downsampled in gamma light, so benchmarking in linar light doesn't really add much.
Artoriuz · Apr 4, 2024 · da30ace · da30ace
1 parent c14c5c4
commit da30ace
Show file tree

Hide file tree

Showing 26 changed files with 83 additions and 388 deletions.
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F16.png b/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F16.png
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F16_LL.png b/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F16_LL.png
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F32.png b/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F32.png
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F32_LL.png b/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F32_LL.png
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F64.png b/blog/images/mpv_upscaling/upsampling/gamma/ArtCNN_C4F64.png
diff --git a/blog/images/mpv_upscaling/upsampling/gamma/benchmark_result.txt b/blog/images/mpv_upscaling/upsampling/gamma/benchmark_result.txt
@@ -1,9 +1,9 @@
-ArtCNN_C4F16, 0.007405294466541064, 33.494250491689336, 0.9831777732629471, 0.9990877844812981
-ArtCNN_C4F16_LL, 0.008438404507710554, 31.92499456733025, 0.9816106317217024, 0.9978683294754817
-ArtCNN_C4F32, 0.006370687670085829, 34.54933571818202, 0.9866138243570441, 0.9992620690770015
-ArtCNN_C4F32_LL, 0.008015850254692626, 32.0374698574192, 0.9839121678866168, 0.997721021786278
+ArtCNN_C4F16, 0.006386227409113109, 34.41525559171364, 0.9860162760141221, 0.9992292633785688
+ArtCNN_C4F32, 0.005481749354546085, 35.61774535315552, 0.989292690612357, 0.9993997206525194
+ArtCNN_C4F64, 0.0052300223292163816, 35.97564044077298, 0.9900907183995443, 0.9994325320232734
 FSRCNNX_x2_16-0-4-1, 0.009745204451887517, 31.720072901827102, 0.9758262429597815, 0.9976227480445303
 FSRCNNX_x2_8-0-4-1, 0.01075761286721094, 31.043815202858593, 0.9705232484760676, 0.9975246837070194
+aoko, 0.02759717046961132, 28.62676289522577, 0.9801801296309367, 0.9820749059658441
 bilinear, 0.02000298653269136, 25.74415445248834, 0.9245016204313588, 0.9888575149131243
 lanczos, 0.015251857860581966, 28.305123482616285, 0.9497111979091768, 0.9962014629617049
 polar_lanczossharp, 0.01611017026027492, 27.874127184654803, 0.9448419907856664, 0.9949491193730922

diff --git a/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F16.png b/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F16.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F16_LL.png b/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F16_LL.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F32.png b/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F32.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F32_LL.png b/blog/images/mpv_upscaling/upsampling/linear/ArtCNN_C4F32_LL.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/FSRCNNX_x2_16-0-4-1.png b/blog/images/mpv_upscaling/upsampling/linear/FSRCNNX_x2_16-0-4-1.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/FSRCNNX_x2_8-0-4-1.png b/blog/images/mpv_upscaling/upsampling/linear/FSRCNNX_x2_8-0-4-1.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/aoko.png b/blog/images/mpv_upscaling/upsampling/linear/aoko.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/benchmark_result.txt b/blog/images/mpv_upscaling/upsampling/linear/benchmark_result.txt
diff --git a/blog/images/mpv_upscaling/upsampling/linear/benchmarker.py b/blog/images/mpv_upscaling/upsampling/linear/benchmarker.py
diff --git a/blog/images/mpv_upscaling/upsampling/linear/bilinear.png b/blog/images/mpv_upscaling/upsampling/linear/bilinear.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/downscaled.png b/blog/images/mpv_upscaling/upsampling/linear/downscaled.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/lanczos.png b/blog/images/mpv_upscaling/upsampling/linear/lanczos.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/polar_lanczossharp.png b/blog/images/mpv_upscaling/upsampling/linear/polar_lanczossharp.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r2.png b/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r2.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r3.png b/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r3.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r4.png b/blog/images/mpv_upscaling/upsampling/linear/ravu-lite-ar-r4.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ravu-zoom-ar-r2.png b/blog/images/mpv_upscaling/upsampling/linear/ravu-zoom-ar-r2.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/ravu-zoom-ar-r3.png b/blog/images/mpv_upscaling/upsampling/linear/ravu-zoom-ar-r3.png
diff --git a/blog/images/mpv_upscaling/upsampling/linear/reference.png b/blog/images/mpv_upscaling/upsampling/linear/reference.png