Commit Graph

187 Commits

Author SHA1 Message Date
76f92964bf Reduce copying in animation step. 2018-03-27 13:22:52 +02:00
e77c68f562 Unify interpolation function. 2018-03-27 11:48:27 +02:00
8b4d29786b Fix an error in the computation of interactions. 2018-03-26 17:18:03 +02:00
0f699c1881 Show layer type in name. 2018-03-26 15:46:36 +02:00
e62ab159a8 Remove useless print statement. 2018-03-26 14:22:50 +02:00
77a274db3b Filter filenames in dialogs. 2018-03-26 14:11:04 +02:00
0c684973a4 Add labels to the file dialogs. 2018-03-26 13:44:26 +02:00
2acff9c822 Fix potential memory leak. 2018-03-26 13:36:34 +02:00
39f3a7507f Allow selection on means file in launcher. 2018-03-26 13:34:39 +02:00
2bd3b5c820 Implement supplying the labels file. 2018-03-26 13:08:58 +02:00
8015b84311 Restructure project sources. 2018-03-26 12:17:52 +02:00
6456ab0612 Search path without new allocations.
At least not for searching. The final result still requires allocating
a string for the final result, but no longer for the intermediates.
2018-03-25 20:35:31 +02:00
55156959d4 Simplify layout. 2018-03-25 19:00:18 +02:00
bb566e2b2d Search for FMRI executable on PATH. 2018-03-25 18:53:31 +02:00
34af03da90 Ensure input is sorted. 2018-03-24 18:03:24 +01:00
6390a9ba1f Reimplement launcher without glade. 2018-03-24 17:32:08 +01:00
f2e2fdd308 Reduce GTK version required. 2018-03-24 16:29:11 +01:00
13015657ab Include GLADE file as literal data. 2018-03-24 16:25:33 +01:00
0851f624db Implement a simple GTK3 launcher. 2018-03-24 15:43:11 +01:00
a0a39c82c7 Simpler build for deinplace tool. 2018-03-24 12:48:56 +01:00
1077f6ab46 Improved error message. 2018-03-23 17:20:39 +01:00
e52e6b7676 Typo in deinplace, somehow. 2018-03-23 16:50:50 +01:00
f8be25adb9 Rewrite the deinplace tool in C++.
More consistency for a better build process.
2018-03-23 16:49:37 +01:00
effa90c178 Use wireframe to give piramids a more 3D feel. 2018-03-22 15:30:30 +01:00
d4ed6e0c0f Don't forget to enable blending. 2018-03-22 15:26:05 +01:00
1492298f2f Smoother rendering. 2018-03-22 15:17:57 +01:00
632c85b2fb Remove magix numbers. 2018-03-22 15:07:13 +01:00
42a5d882d6 Fix typo in layer name. 2018-03-20 15:35:29 +01:00
229b7c8004 Improved layer name -> type mappings. 2018-03-20 14:22:15 +01:00
c0cf5e6ff0 Improve CMake structure. 2018-03-18 21:26:55 +01:00
512c4e8c95 Tweak animation step a little.
Animation now starts slightly after the start and ends slightly before
the end. This prevents a weird flicker effect in the image like layers.
2018-03-16 14:37:37 +01:00
40a60c02bb Improve CMakeLists.txt 2018-03-16 14:31:48 +01:00
c814d274d4 Unify drawing logic into glUtils. 2018-03-16 14:25:14 +01:00
c447a496fd Improved visualisation of ReLU on images. 2018-03-13 15:06:48 +01:00
74ecd42abe Show normalization rates in visualisation. 2018-03-13 15:06:40 +01:00
93d4cb4df0 Remove redundant code. 2018-03-13 14:28:00 +01:00
e21c7b33e8 Start implementing visualisation for LRN layers. 2018-03-12 17:07:51 +01:00
f7f3eb5196 Add a warning for networks with in-place layers. 2018-03-12 13:38:46 +01:00
38d28c9b7c Implement a simple coloring for ReLU layers.
Note: might not actually work.
2018-03-06 16:11:08 +01:00
7963562717 Create extendable coloring for interactions. 2018-03-06 14:50:02 +01:00
18306e45b8 Implement basic visualisation for pooling layers. 2018-03-01 16:16:21 +01:00
0b8c05ded9 Unify texture configuration. 2018-02-27 16:04:38 +01:00
7fdce48259 More generalized code.
MultiImageLayer and FlatLayer now use the same code for arranging nodes,
rather than then same logic twice.
2018-02-27 15:32:38 +01:00
2dec364b1b Add debug info in the top left corner. 2018-02-26 15:39:53 +01:00
604d302627 Implement a specialized input visualisation. 2018-02-26 14:37:02 +01:00
17d4e07025 Implement a small texture wrapper.
Allows for RAII use of OpenGL textures.
2018-02-25 19:03:10 +01:00
23ae9717ea Slightly nicer animations. 2018-02-23 17:59:15 +01:00
63b671f54d Use caffe multiplication for efficiency. 2018-02-23 16:05:34 +01:00
a8857faa24 Improve performance of animation layers. 2018-02-23 15:51:45 +01:00
99cf0740fe Improve convolutional layer rendering.
Now all n individual images of w×h are packaged as a w×(n×h) image, with
the texture coordinates properly mapped so the individual rectangles
still get the proper texture.

This optimization causes slight glitches around the edges of the tiles,
but is so much faster that I think it's worth it.
2018-02-23 15:19:30 +01:00