08550ac856
fix H128 after refactoring
2019-01-04 18:28:07 +04:00
69d5b1e662
fix UFM
2019-01-03 23:02:38 +04:00
7b70e7eec1
added new model U-net Face Morpher.
...
removed AVATAR - useless model was just for demo
removed MIAEF128 - use UFM insted
removed LIAEF128YAW - use model option sort by yaw on start for any model
All models now ask some options on start.
Session options (such as target epoch, batch_size, write_preview_history etc) can be overrided by special command arg.
Converter now always ask options and no more support to define options via command line.
fix bug when ConverterMasked always used not predicted mask.
SampleGenerator now always generate samples with replicated border, exclude mask samples.
refactorings
2019-01-02 17:26:12 +04:00
46bda1d683
RecycleGAN : set minimum bs to 10
2018-12-28 23:42:05 +04:00
474fff248f
RecycleGAN fixes
2018-12-28 23:13:29 +04:00
f8824f9601
refactoring. Added RecycleGAN for testing.
2018-12-28 19:38:52 +04:00
8a223845fb
refactoring
2018-12-24 13:45:40 +04:00
0d7387165a
added --cpu-only mode for all stages. Upd readme. Added requirements-cpu.txt
2018-12-21 16:54:22 +04:00
ffddfeabdb
added converter --seamless-erode-mask-modifier . This mask used to define area for opencv seamless cloning function. So if you erode it little bit , you can decrease flickering effect on some frames.
2018-12-20 13:50:38 +04:00
9926dc626a
refactorings, improved sort by hist-dissim
2018-12-20 12:43:00 +04:00
b2d28e0b95
added converter option --use-predicted-mask . By default converter uses model's predicted mask, but now you can use DST mask.
2018-12-19 23:28:48 +04:00
98ee1bb339
Added option hist-match threshold, used for decrease artifacts of hist-match.
...
Fix terminating converter thread in debug mode.
Added DeepFaceLab converter overview to readme.
2018-12-14 10:40:46 +04:00
016791bb6f
fix output scale modifier for expanding face
2018-12-13 00:57:55 +04:00
1bd3a8e0c1
fix error on small mask region size in converter,
...
error catching on convert face
2018-12-12 20:48:30 +04:00
5c43f4245e
transfercolor via lab converter now implemented by tensorflow-cpu, which is x2 faster than skimage.
...
We cannot use GPU for lab converter in converter multiprocesses, because almost all VRAM ate by model process, so even 300Mb free VRAM not enough for tensorflow lab converter.
Removed skimage dependency.
Refactorings.
2018-12-01 12:11:54 +04:00
6f0d38d171
converter added --final-image-color-degrade-power - Degrades colors of final image to hide face problems. Valid range [0..100]
2018-11-30 20:08:30 +04:00
cef6710d24
enhanced masked hist match
2018-11-28 21:07:26 +04:00
64c3e57f1c
added option to converter --output-face-scale-modifier
2018-11-28 20:38:48 +04:00
6bbb8ad59b
Added option --alpha and --transfercolor
2018-11-17 18:14:17 +03:00
58d9b4ae09
fix avatar multigpu mode
2018-06-07 12:18:42 +04:00
adc1e701de
increased H64 decoder size for 4GB+
2018-06-07 10:06:24 +04:00
4aaac5e42e
AVATAR fixes
2018-06-06 18:43:11 +04:00
612ef5155e
increased speed of train on H64, H128 models.
2018-06-06 18:40:36 +04:00
02331c301b
min vram -0.5gb for all models
2018-06-05 16:23:43 +04:00
d5f5a35945
min 1.5gb for H64
2018-06-05 16:20:40 +04:00
d23c8ecfb9
AVATAR now 256
2018-06-05 09:17:25 +04:00
6bd5a44264
initial
2018-06-04 17:12:43 +04:00
iperov