0f42348 | DeadMetaler | 28 September 2016, 00:06:07 UTC | Update noiseprofiles.json | 28 September 2016, 00:06:07 UTC |
1268781 | houz | 27 September 2016, 22:10:13 UTC | Merge pull request #1269 from TuTPhoto/master Fixed rather all fuzzy strings | 27 September 2016, 22:10:13 UTC |
08b699a | Maurizio Paglia | 27 September 2016, 21:33:53 UTC | Fixed rather all fussy strings | 27 September 2016, 21:33:53 UTC |
3b56208 | Roman Lebedev | 27 September 2016, 18:23:51 UTC | Merge remote-tracking branch 'upstream/pr/1265' * upstream/pr/1265: Default tiling_callback() changes. | 27 September 2016, 18:23:51 UTC |
331c77c | Roman Lebedev | 27 September 2016, 14:08:45 UTC | .travis.yml: enable very noisy mail notifications to new -ci ML | 27 September 2016, 14:08:53 UTC |
c56ccfa | Roman Lebedev | 27 September 2016, 13:46:32 UTC | Nikon 1 J3: appears to actually have whitelevel of 4095. While dng converter produces wl of 3300 for iso <200, visual inspection of such an image shows no purple highlight with whitelevel of 4095. | 27 September 2016, 13:46:37 UTC |
a73131c | Roman Lebedev | 27 September 2016, 12:56:33 UTC | Nikon 1 V2: appears to actually have whitelevel of 4095. While dng converter produces wl of 3300 for iso <200, visual inspection of such an image shows no purple highlight with whitelevel of 4095. Closes #11171. | 27 September 2016, 12:57:45 UTC |
9bfe0ae | Tobias Ellinghaus | 27 September 2016, 10:55:26 UTC | .travis.yml: Make the bot not join/part That gives too many useless messages in IRC | 27 September 2016, 10:55:26 UTC |
8f58a98 | Roman Lebedev | 19 September 2016, 20:31:34 UTC | Default tiling_callback() changes. 1. All operations that work with mosaiced data should respect patter size! 2. Also attempt to automatically handle simple case of buffer size change. NOTE: this does not attempt to guess additional required buffer sizes. 3. Attempt to set overlap (to 4) based on flag IOP_FLAGS_TILING_FULL_ROI | 27 September 2016, 10:51:24 UTC |
4c2dbde | Tobias Ellinghaus | 27 September 2016, 10:40:41 UTC | .travis.yml: Notify IRC about build results | 27 September 2016, 10:40:41 UTC |
21d6ae9 | Roman Lebedev | 27 September 2016, 10:27:46 UTC | .travis.yml: an attempt to fix build matrix | 27 September 2016, 10:27:50 UTC |
f9faebc | Roman Lebedev | 27 September 2016, 10:17:46 UTC | .travis.yml: an attempt to parallelize and speed-up | 27 September 2016, 10:17:55 UTC |
733c67a | Roman Lebedev | 27 September 2016, 09:54:10 UTC | README.md: use better links | 27 September 2016, 09:54:14 UTC |
a7876f2 | Roman Lebedev | 27 September 2016, 09:33:22 UTC | Move docker image into new DockerHub org, add icons to README.md Also, add a comment warning about no auto-rebuild. | 27 September 2016, 09:33:49 UTC |
d6a0fa1 | Roman Lebedev | 27 September 2016, 08:36:31 UTC | Add Dockerfile, .travis.yml; Enable CI. | 27 September 2016, 08:37:18 UTC |
46a8df5 | Pascal Obry | 26 September 2016, 15:50:44 UTC | Fix OpenCL compilation error. The message was: BUILD LOG: <kernel>:1912:3: error: subscripted access is not allowed for OpenCL vectors pixel[c] = 0.0; ^ ~ | 26 September 2016, 15:50:44 UTC |
a4d29b0 | johannes hanika | 26 September 2016, 09:23:24 UTC | Merge remote-tracking branch 'origin/pr/1241' into tmp | 26 September 2016, 09:23:24 UTC |
63dbcb3 | Tobias Ellinghaus | 25 September 2016, 21:23:19 UTC | Quick tagging searches anywhere, not just at start Fixes #9946 Fixes #10853 Fixes #10139 | 25 September 2016, 21:23:19 UTC |
83c70b8 | Roman Lebedev | 25 September 2016, 21:12:41 UTC | CMake: cleanup RPATH handling a little. There is no such properties like CMAKE_BUILD_WITH_INSTALL_RPATH and CMAKE_INSTALL_RPATH_USE_LINK_PATH, those are global vars :) And all those properties are initialized with some global vars first. Since we always set those props to the same values, let's try to just set global variables. | 25 September 2016, 21:18:10 UTC |
4129a30 | Tobias Ellinghaus | 25 September 2016, 13:58:10 UTC | [tags] Allow to import/export tags This is useful when using a closed vocabulary for tagging images. A way to disable creating new tags is needed still, but this at least allows to get tags into dt. There are some free tag lists available on the internet, for example - Some CC-BY-SA licensed lists: https://lightroom-keyword-list-project.blogspot.de/ - Not sure about the license: http://workshops.bigsunphotography.com/news/28-free-lightroom-keyword-library-download Google might find more. Just make sure that the txt file uses the Lightroom format as described here: http://www.photo-keywords.com/keyword-list-format.php | 25 September 2016, 13:58:10 UTC |
bab1824 | Tobias Ellinghaus | 25 September 2016, 11:45:48 UTC | Remove references to tagxtag in comments | 25 September 2016, 11:45:48 UTC |
64303b8 | Tobias Ellinghaus | 25 September 2016, 11:38:51 UTC | Merge remote-tracking branch 'origin/pr/1247' | 25 September 2016, 11:38:51 UTC |
81f2e11 | Tobias Ellinghaus | 24 September 2016, 23:08:18 UTC | Fix theoretical off-by-one error This part of the code can't be reached at the moment, but I guess it should be correct nevertheless. At least I hope that this is right. | 24 September 2016, 23:08:18 UTC |
0610eca | Roman Lebedev | 24 September 2016, 20:50:16 UTC | dt_mipmap_buffer_t: add iscale. It only makes sense for MIP_F and / or MIP_FULL. For MIP_FULL, it is 1.0f; else it is MIP_FULL/MIP_F scale factor (> 1.0f) (for < MIP_F it may be a garbage, but i tried to set it to 1.0f if there is any data, and to 0.0f if there is no data) Follow-up to 8d18f6f56f2073bc11a44465abb7163cd87f4445. Refs #10566, #10595. | 24 September 2016, 20:50:21 UTC |
ad65647 | Roman Lebedev | 24 September 2016, 18:18:35 UTC | Canon 700D: better levels based on provided samples. Thanks to Dutch_Wolf for the samples. | 24 September 2016, 18:21:30 UTC |
2f5d21e | Roman Lebedev | 24 September 2016, 16:11:28 UTC | dngmeta.rb: fixes. Mostly handling of duplicate information. For canon i now try to group by whitelevel, and then just pick the highest blacklevel. Since for canon rawspeed always calculates blacklevel, it is ok. | 24 September 2016, 18:20:26 UTC |
66fa08d | Tobias Ellinghaus | 24 September 2016, 13:08:24 UTC | It's "exposure", not "exposition" | 24 September 2016, 13:08:24 UTC |
da651ae | Roman Lebedev | 24 September 2016, 12:54:46 UTC | Nikon D4: proper whitelevels for all the modes, enable camera. Thanks to Christian Deckelmann for full sample set (each mode per each base iso). | 24 September 2016, 12:55:15 UTC |
b43922f | Roman Lebedev | 23 September 2016, 20:45:16 UTC | CMake: move cmake_policy(CMP0011) into top-level cmakelists.txt There is no reason to keep it only for src/ See https://cmake.org/cmake/help/v3.0/policy/CMP0011.html | 23 September 2016, 20:49:36 UTC |
bebc418 | Roman Lebedev | 23 September 2016, 20:40:48 UTC | IOP CMakeLists: forgot to remove cmake_minimum_required leftover | 23 September 2016, 20:49:36 UTC |
f0398cf | Roman Lebedev | 23 September 2016, 20:29:09 UTC | create_release.sh: manually do XZ compression. I did compare every compression setting with the tar-produced .tar.xz, and manually compressing with options -9e results in the smallest archive, with the 80% size of the one produced by tar directly. This of course comes with speed and memory (de)compression penalty, but since the archive is created just once, it should be fine. | 23 September 2016, 20:49:35 UTC |
07a88a9 | Roman Lebedev | 23 September 2016, 20:23:57 UTC | create_release.sh: fix version_gen.c generation I broke it when i changed it from simple echo to script call. Also, makes it possible to run this script from any directory, not just root | 23 September 2016, 20:49:35 UTC |
faaff23 | Roman Lebedev | 23 September 2016, 17:03:31 UTC | CMake: unconditionaly enable CMAKE_EXPORT_COMPILE_COMMANDS. It is an option since 2.8.5. It is required for KDevelop 5, which i use, and i will not be bothered to pass -DCMAKE_EXPORT_COMPILE_COMMANDS=ON to cmake each time :) | 23 September 2016, 17:26:37 UTC |
8826ace | Pascal Obry | 23 September 2016, 16:58:07 UTC | Update French translation. | 23 September 2016, 16:58:07 UTC |
2ba80b2 | Tobias Ellinghaus | 23 September 2016, 15:17:21 UTC | Mark a string for translation ... forgotten in the last commits. Thanks Roman for noticing. | 23 September 2016, 15:17:21 UTC |
f178ae0 | Tobias Ellinghaus | 23 September 2016, 15:03:22 UTC | Treat non-db files like corrupted db files at init | 23 September 2016, 15:03:22 UTC |
5f2d005 | Tobias Ellinghaus | 23 September 2016, 14:55:01 UTC | Fix #11170: Handle corrupted library gracefully | 23 September 2016, 14:55:01 UTC |
d1b32e0 | Ulrich Pegelow | 23 September 2016, 13:40:54 UTC | ashift: add sanity check for degenerate cases In rare cases automatic parameter fitting can lead to insane results with so strong image distortions that the required image area is drastically higher than the original one (a factor of above 20 has been observed). As the memory requirements will drastically increase this could even bring the system down. | 23 September 2016, 13:40:54 UTC |
f583eda | Roman Lebedev | 22 September 2016, 17:41:44 UTC | Rawprepare iop: process_sse2(): raw mosaic: actually don't clamp :( Was missed in 76832e23a6df2237683ff72c906ad7aabef49bf9. | 22 September 2016, 17:42:05 UTC |
f68ef08 | Roman Lebedev | 22 September 2016, 17:20:13 UTC | Invert IOP: X-Trans float mosaic: SSE Finally, 2 years after saying i'll look into it :) The only difference from Bayer SSE code is that X-Trans sensor is 6x6, and SSE processes 4 pixels at once. Since fetching colors (FCxtrans) each time would kill performance, we need to process 12 pixels at once. Why 12? Because that is the smallest x for which the following two equations are true: 0) x mod 4 == 0 (SSE requirement) 1) x mod 6 == 0 (X-Trans sensor size requirement) | 22 September 2016, 17:25:46 UTC |
b09bcde | Roman Lebedev | 22 September 2016, 17:02:55 UTC | Temperature IOP: X-Trans float mosaic: SSE Finally, 2 years after saying i'll look into it :) The only difference from Bayer SSE code is that X-Trans sensor is 6x6, and SSE processes 4 pixels at once. Since fetching colors (FCxtrans) each time would kill performance, we need to process 12 pixels at once. Why 12? Because that is the smallest x for which the following two equations are true: 0) x mod 4 == 0 (SSE requirement) 1) x mod 6 == 0 (X-Trans sensor size requirement) Profiling results are inconclusive. New code is neither statistically-significantly worse nor statistically-significantly better than the old one. But -d perf says this new code is just a tiny bit better. | 22 September 2016, 17:02:55 UTC |
73c9faf | Roman Lebedev | 04 September 2016, 16:11:13 UTC | Rawoverexposed iop: explain why we account for wb. | 21 September 2016, 20:19:55 UTC |
ca05f9c | Roman Lebedev | 05 August 2016, 19:47:18 UTC | Rawoverexposed iop: blind attempt at opencl kernel for mark cfa mode | 21 September 2016, 20:19:54 UTC |
7717fa5 | Roman Lebedev | 05 August 2016, 19:47:04 UTC | Rawoverexposed iop: blind attempt at opencl kernel for mark solid mode | 21 September 2016, 20:19:54 UTC |
994c844 | Roman Lebedev | 04 September 2016, 16:13:28 UTC | Rawoverexposed iop: blind attempt at opencl kernel for false color mode | 21 September 2016, 20:19:54 UTC |
1fd6104 | Roman Lebedev | 04 September 2016, 16:09:48 UTC | Rawoverexposed: second attempt, after overexposed, sample MIP_FULL. Much better, configurable since it works with 4 channels. | 21 September 2016, 20:19:54 UTC |
99cf25e | Roman Lebedev | 06 August 2016, 15:16:01 UTC | Temperature iop: store wb coeffs also in dsc | 21 September 2016, 20:19:54 UTC |
f5ee08a | Roman Lebedev | 02 August 2016, 12:28:24 UTC | Add new raw overexposure indication iop Right now it is located right before highlight reconstruction iop, so it operates one the raw data, before demosaicing. As a result, for the areas with "clipped" colors, it produces flat areas with false color. 1. Not configurable. 2. Will be affected by all the iops that are after this module. | 21 September 2016, 20:18:20 UTC |
034c663 | Ulrich Pegelow | 21 September 2016, 20:10:04 UTC | opencl: add whitebalance kernel for x-trans images | 21 September 2016, 20:10:04 UTC |
a0ad7fb | Roman Lebedev | 21 September 2016, 17:43:32 UTC | Pixelpipe cache: dt_dev_pixelpipe_cache_flush(): do poison. Just like dt_dev_pixelpipe_cache_invalidate(). | 21 September 2016, 17:43:32 UTC |
370e07d | Roman Lebedev | 21 September 2016, 17:39:57 UTC | Pixelpipe cache: dt_dev_pixelpipe_cache_reweight(): don't poison. Don't think reweight should poison the buffer. That poisoning was supposed to go into dt_dev_pixelpipe_cache_flush() | 21 September 2016, 17:39:57 UTC |
4e8a98a | Roman Lebedev | 21 September 2016, 17:20:07 UTC | Oops, fix build without ASan :) | 21 September 2016, 17:20:07 UTC |
8350797 | Roman Lebedev | 21 September 2016, 17:09:46 UTC | dt_dev_pixelpipe_process_rec(): don't memset() more than the size you asked for | 21 September 2016, 17:12:08 UTC |
2890992 | Roman Lebedev | 21 September 2016, 16:57:59 UTC | Pixelpipe cache: manually (un)poison memory We keep buffers and reuse them to not re-allocate each time. While this does save us from malloc(3) overhead, this also shadows the actual size that e think is usable from AddressSanitizer. I.e. we first allocate buf of size 4, use it, and then re-use it but with size 2. And if there is some code that overflows it, we may not know about it. With this hacky manual (un)poisoning, hopefully, it will be easier to catch buffer overflows. | 21 September 2016, 17:11:53 UTC |
d9dda34 | Roman Lebedev | 21 September 2016, 15:46:23 UTC | config.h: add macros for manual ASan's memory (un)poisoning In some places we keep buffers and reuse them to not re-allocate each time. While this does save us from malloc(3) overhead, this also shadows the actual size that we think is usable from AddressSanitizer. I.e. we first allocate buf of size 4, use it, and then re-use it but with size 2. And if there is some code that oveflows it, we may not know about it. | 21 September 2016, 17:11:29 UTC |
afaca87 | Ulrich Pegelow | 21 September 2016, 15:41:29 UTC | opencl: improve readability of profiling data for multi-GPU environments | 21 September 2016, 15:41:29 UTC |
1e4164a | Roman Lebedev | 21 September 2016, 14:53:33 UTC | Pipe: properly(?) pass buffer dsc this time. The fix seems almost too trivial to be correct. Refs. #11168 (hopefully fixes) Refs. #11149 (hopefully does not reintroduce) | 21 September 2016, 15:04:57 UTC |
531cfee | Roman Lebedev | 21 September 2016, 13:22:55 UTC | Pipe: get_output_format(): if first input, always use provided dsc. Buf it not raw, override processed_maximum to 1.0 | 21 September 2016, 15:04:57 UTC |
485169a | Roman Lebedev | 21 September 2016, 13:17:16 UTC | Rawprepare iop: the output is normalized so 1.0f is processed_maximum | 21 September 2016, 15:04:56 UTC |
30296a9 | Tobias Ellinghaus | 21 September 2016, 11:27:29 UTC | Point to new ML archive and use https links | 21 September 2016, 11:27:29 UTC |
1f5cfff | Pascal Obry | 20 September 2016, 19:37:54 UTC | If there is no local-copy there is nothing to sync. Check whether the image has a local copy or not. If not we exit as there is really nothing to sync and so we avoid a spurious error reported when an image has been removed/renamed outside of dt. For #10725. | 20 September 2016, 20:00:41 UTC |
4bc1cab | Roman Lebedev | 20 September 2016, 18:59:09 UTC | Imageop_math, mipmap: dead code removal | 20 September 2016, 18:59:11 UTC |
069118e | Roman Lebedev | 20 September 2016, 18:35:30 UTC | Populate ugly FILTERS_ARE_4BAYER() with more values. Fixes #11166. The problem is that it may compare those values with arbitrary shifted pattern. So for every 4-color CFA this needs to contain all 4 different variants. This is absolutely horrible. | 20 September 2016, 18:40:14 UTC |
f22159e | Roman Lebedev | 20 September 2016, 11:00:06 UTC | Demosaic: allow passthrough (monochrome) method for all sensors. I need it for guess-crop.py script to guess crop (duh) of camera for rawspeed's cameras.xml Resolves #1141. | 20 September 2016, 18:33:55 UTC |
bf26d00 | Roman Lebedev | 20 September 2016, 15:01:42 UTC | Merge remote-tracking branch 'upstream/pr/1268' * upstream/pr/1268: Panasonic DMC-GX85/DMC-GX80: Basic support added, fixes #11141 | 20 September 2016, 15:02:28 UTC |
fc9f3ad | Ger Siemerink | 20 September 2016, 13:05:02 UTC | Updated Dutch translation | 20 September 2016, 13:05:02 UTC |
cae6742 | Roman Lebedev | 19 September 2016, 20:26:05 UTC | dt_dev_pixelpipe_process_rec(): input_format shenanigans. Fixes. #11149 Due to the block starting with comment // we're looking for the full buffer we may have wrong buffer description. For now, we can just workaround it by calling input_format() iop callback. FIXME: actually understand how to fix this properly! | 19 September 2016, 20:26:31 UTC |
b83264f | Ulrich Pegelow | 18 September 2016, 18:05:35 UTC | opencl: always use blocking memory transfer host<->device NVIDIA devices in combination with newer drivers (after 340.xx) show a marked slow-down when non-blocking memory transfer is used. Reason is not known. Tests show that blocking memory transfer brings back the expected performance. | 18 September 2016, 18:05:35 UTC |
b64da29 | johannes hanika | 18 September 2016, 17:58:12 UTC | Merge remote-tracking branch 'origin/pr/1263' into tmp | 18 September 2016, 17:58:12 UTC |
d165261 | Roman Lebedev | 18 September 2016, 17:55:40 UTC | dt_iop_clip_and_zoom_mosaic_half_size*(): don't forget about offset. Fixes preview pipe for cameras with crop which is not multiple of 2, like canon 5d mk2. | 18 September 2016, 17:56:32 UTC |
61712bb | Tobias Ellinghaus | 18 September 2016, 17:29:00 UTC | Always show rendering intent selector in lt We don't use our fast path for color managing lighttable so there is no need to disable the intent setting based on the export option. Related to #11161. | 18 September 2016, 17:29:00 UTC |
6f84c34 | Maurizio Paglia | 17 September 2016, 16:45:38 UTC | [usermanual] Update Italian translation This is basically a completely new translation, the old one was WAY behind. (cherry picked from commit 634961c1f29db0770773d8dceb01e33df714e654) | 17 September 2016, 21:14:53 UTC |
588f4d0 | Boucman | 17 September 2016, 12:56:11 UTC | lua : allow to add any widget in lua preference | 17 September 2016, 12:56:11 UTC |
61651df | Stefan Schöfegger | 11 September 2016, 12:57:27 UTC | Panasonic DMC-GX85/DMC-GX80: Basic support added, #11141 GX7mkII should be the same camera, but I could not find any samles to see how it is written in exif info. So it could not be added as alias. | 16 September 2016, 16:59:03 UTC |
07ed916 | Pascal Obry | 16 September 2016, 15:34:13 UTC | Update French translation. | 16 September 2016, 15:34:13 UTC |
05fd3dc | Roman Lebedev | 15 September 2016, 22:12:57 UTC | dt_dev_pixelpipe_set_input(): just use get_output_format(). Fixes #11149 I bet that would not have been an issue with my MIP_F rework ^^ (PR darktable-org/darktable#1263) | 15 September 2016, 22:15:14 UTC |
3966d04 | Roman Lebedev | 15 September 2016, 13:14:58 UTC | LICENSE: move it to where it belongs. For one, it makes github actually detect and show the license. | 15 September 2016, 13:17:56 UTC |
833001e | Roman Lebedev | 15 September 2016, 12:45:35 UTC | CMake 3.0 default behaviour change: don't unset PROJECT_VERSION/VERSION | 15 September 2016, 12:46:45 UTC |
4f4e89a | Roman Lebedev | 15 September 2016, 12:20:10 UTC | Merge pull request #1264 from mazhe/master Rawspeed: use a non-deprecated way of getting OSX version | 15 September 2016, 12:20:10 UTC |
71ff823 | Matthieu Volat | 15 September 2016, 11:54:37 UTC | When reading osx version via sysctl, ensure the result won't make darktable crash Validate it's a string, and that has the right format. | 15 September 2016, 11:54:37 UTC |
6bb1d05 | Matthieu | 15 September 2016, 10:36:38 UTC | Merge remote-tracking branch 'darktable-org/master' | 15 September 2016, 10:36:38 UTC |
82a45ae | Matthieu | 15 September 2016, 10:36:10 UTC | Merge branch 'master' of github.com:mazhe/darktable | 15 September 2016, 10:36:10 UTC |
f6fec2d | Matthieu Volat | 15 September 2016, 10:34:43 UTC | Use a non-deprecated way of getting OSX version Gestalt stuff from Carbon is deprecated, use a more stable sysctl call to get OSX version, but keep the previous version format. | 15 September 2016, 10:34:43 UTC |
72aac66 | Roman Lebedev | 14 September 2016, 21:21:09 UTC | X-Trans float: implement a bit better downscaling (just like the bayer) prone to confetti colours, does not do sub-sampling. But at least not as pixellated :) Simple adaptation from dt_iop_clip_and_zoom_demosaic_third_size_xtrans_f() | 14 September 2016, 21:21:09 UTC |
f4bd099 | Roman Lebedev | 14 September 2016, 21:20:24 UTC | X-Trans uint16_t: implement a bit better downscaling (just like the bayer) prone to confetti colours, does not do sub-sampling. But at least not as pixellated :) FIXME: does it have to do that whitelevel stuff? Simple adaptation from dt_iop_clip_and_zoom_demosaic_third_size_xtrans() | 14 September 2016, 21:20:24 UTC |
dfea4ed | parafin | 14 September 2016, 20:56:53 UTC | mac: drop support for OS X 10.6 | 14 September 2016, 20:56:53 UTC |
870783b | parafin | 14 September 2016, 20:17:53 UTC | cmake: don't assume C and C++ compilers are the same case in point is mac package (cherry picked from commit 8009bde930747e9a725419e61caa91163fbfa556) | 14 September 2016, 20:24:52 UTC |
5368700 | parafin | 14 September 2016, 20:24:47 UTC | mac: bump gcc version to 5 | 14 September 2016, 20:24:47 UTC |
e58619d | parafin | 14 September 2016, 20:22:54 UTC | cmake: enable CMP0025 for mac so we can distinguish Apple's clang from ordinary and bump cmake version requirement to 3.0 for it | 14 September 2016, 20:22:54 UTC |
dbc3178 | Roman Lebedev | 12 September 2016, 17:10:04 UTC | Imageop math: subpixel sampling: fix lower right 2x2 block calculation | 14 September 2016, 18:35:25 UTC |
91d54ac | Roman Lebedev | 12 September 2016, 16:33:40 UTC | _init_f(): correct scaling factor calculation Now we only do simple&stupid binning for X-Trans. FIXME: implement proper downscaling with sub-pixel sampling for X-Trans? | 14 September 2016, 18:35:25 UTC |
6c2ad3f | Roman Lebedev | 12 September 2016, 16:22:08 UTC | Bayer float: implement proper downscaling, with sub-pixel sampling. Should help with stair-casing? Simple adaptation from dt_iop_clip_and_zoom_mosaic_half_size*() | 14 September 2016, 18:35:25 UTC |
7b2cea1 | Roman Lebedev | 12 September 2016, 16:26:31 UTC | Bayer uint16_t: implement proper downscaling, with sub-pixel sampling. Should help with staircasing? Simple adaptation from dt_iop_clip_and_zoom_demosaic_half_size_f*() | 14 September 2016, 18:35:25 UTC |
3f8ca97 | Roman Lebedev | 12 May 2016, 18:04:13 UTC | Mipmap buffer: pre_monochrome_demosaiced be gone. | 14 September 2016, 18:35:25 UTC |
13a207b | Roman Lebedev | 15 May 2016, 11:07:30 UTC | dt_dev_pixelpipe_uses_downsampled_input() is no more. We do need to keep 4f part of rawprepare because of sRAW's and DNG's. | 14 September 2016, 18:35:25 UTC |
9cf22d5 | Roman Lebedev | 08 September 2016, 13:56:20 UTC | _init_f(): take into account that MIP_F is 4 channels | 14 September 2016, 18:35:25 UTC |
341c574 | Roman Lebedev | 06 May 2016, 15:21:05 UTC | Imageop, mipmap cache: implement pixel binning for CFA images. [float] | 14 September 2016, 18:35:25 UTC |
1886139 | Roman Lebedev | 04 May 2016, 13:47:02 UTC | Imageop, mipmap cache: implement pixel binning downscaling for CFA images. Unlike others, this does not demosaic, and still outputs uint16_t, so we will be able to use full pixelpipe with demosaicing even for preview and thumbnail pipes. As a result, at the very least, the histogram will present more useful information about highlights. | 14 September 2016, 18:35:25 UTC |
8444a39 | Roman Lebedev | 01 May 2016, 16:22:41 UTC | dt_dev_pixelpipe_uses_downsampled_input(): just disable for now. The idea is to drop it. | 14 September 2016, 18:35:25 UTC |
f397ede | Roman Lebedev | 15 May 2016, 10:40:42 UTC | dt_iop_clip_and_zoom_demosaic_half_size_f(): don't care about whitelevel. | 14 September 2016, 18:35:24 UTC |