 Phil Wang
|
532178a6a3
|
assume when sampling all batch samples are at the same time, and do not noise for the last time step
v0.16.4
|
2022-06-07 16:12:44 -07:00 |
|
|
Phil Wang
|
3bbb6ebf16
|
get working version of gaussian diffusion with continuous time (only beta linear schedule for now, but will eventually contain alpha cosine schedule as well as parameterized, learned monotonic MLP)
v0.16.3
|
2022-06-07 15:59:29 -07:00 |
|
|
Phil Wang
|
6b93fa48f6
|
fix comment
|
2022-06-06 17:30:46 -07:00 |
|
|
Phil Wang
|
a291da5098
|
bring back linear noise schedule, but default to cosine
v0.16.1
|
2022-05-27 19:13:05 -07:00 |
|
|
Phil Wang
|
e5a18bb25c
|
switch over to film like conditioning, used by both openai and google at this point
v0.16.0
|
2022-05-24 23:47:34 -07:00 |
|
|
Phil Wang
|
fc8e4547aa
|
higher default learning rate
0.15.7
|
2022-05-16 13:39:55 -07:00 |
|
|
Phil Wang
|
cae9f4a71f
|
whoops
0.15.6
|
2022-05-14 13:59:21 -07:00 |
|
|
Phil Wang
|
91f03fb88b
|
optimize for simplicity and clarity - researcher does not need to worry about normalizing and unnormalizing now
0.15.4
|
2022-05-14 11:38:43 -07:00 |
|
|
Phil Wang
|
60128257c5
|
use tqdm pbar during training
0.15.3
|
2022-05-13 20:25:49 -07:00 |
|
|
Phil Wang
|
cf6db71985
|
add gaussian diffusion where model predicts both noise and x_start, with a learned weighting between the two (experimental)
0.15.2
|
2022-05-13 13:56:32 -07:00 |
|
|
Phil Wang
|
84ebb9ad13
|
offer predict_x0 objective
0.15.0
|
2022-05-13 10:15:54 -07:00 |
|
|
Phil Wang
|
caa5af170d
|
final cleanup
0.14.3
|
2022-05-12 13:58:16 -07:00 |
|
|
Phil Wang
|
55c658b967
|
cleanup unused
|
2022-05-12 11:52:06 -07:00 |
|
|
Phil Wang
|
e0f26677d6
|
make sure predicted mean is actually detached for all of the kl loss calculations
0.14.2
|
2022-05-12 11:12:34 -07:00 |
|
|
Phil Wang
|
e147839d74
|
make sure to clip when sampling from gaussian diffusion with learned variance
0.14.1
|
2022-05-12 10:08:53 -07:00 |
|
|
Phil Wang
|
62e8490385
|
complete the gaussian diffusion with hybrid loss (learned variance) as in the improved ddpm paper
0.14.0
|
2022-05-12 08:54:47 -07:00 |
|
|
Phil Wang
|
d412d8816b
|
first pass at ddpm with learned variance
|
2022-05-11 17:38:29 -07:00 |
|
|
Phil Wang
|
402b7c26df
|
calculate noise schedule with float64 for numerical accuracy
0.12.1
|
2022-05-10 15:23:34 -07:00 |
|
|
Phil Wang
|
09613a40f3
|
cleanup
|
2022-05-07 05:47:21 -07:00 |
|
|
Phil Wang
|
c6966ae95a
|
Merge pull request #24 from kashif/patch-1
updated citation in README
|
2022-05-07 05:32:51 -07:00 |
|
|
Kashif Rasul
|
73591cf1ad
|
updated citation in README
|
2022-05-07 11:23:45 +02:00 |
|
|
Phil Wang
|
989f0fcb8e
|
remove convnext blocks, they do not work well, validated in video diffusion repository
0.12.0
|
2022-05-05 07:03:55 -07:00 |
|
|
Phil Wang
|
84731bb03d
|
groupnorm groups should be actually configurable
0.11.2
|
2022-05-04 10:38:29 -07:00 |
|
|
Phil Wang
|
c6ecca555b
|
allow for configuring expansion factor in convnext
0.11.1
|
2022-05-04 10:33:23 -07:00 |
|
|
Phil Wang
|
1f5c233072
|
bring back resnet blocks, make convnext blocks an experimental option
0.11.0
|
2022-05-04 10:30:09 -07:00 |
|
|
Phil Wang
|
de378158e5
|
readme
|
2022-05-01 13:16:06 -07:00 |
|
|
Phil Wang
|
e274fb305a
|
give an initial conv
0.10.1
|
2022-05-01 08:49:38 -07:00 |
|
|
Phil Wang
|
f39b3b1d3f
|
make sure time embedding dimension is kept at 4 x dimension (thanks @borisdayma)
0.10.0
|
2022-04-29 14:55:12 -07:00 |
|
|
Phil Wang
|
782c904d3b
|
fix cosine beta schedule, thanks to @Zhengxinyang
0.9.2
|
2022-04-19 20:51:50 -07:00 |
|
|
Phil Wang
|
71953ebd22
|
fix bug, thanks to @jihoonerd
0.9.1
|
2022-04-15 06:37:31 -07:00 |
|
|
Phil Wang
|
0b8cdb4c8b
|
remove outdated apex in favor of native pytorch AMP
0.9.0
|
2022-04-13 08:59:18 -07:00 |
|
|
Phil Wang
|
e504e0e554
|
cleanup
|
2022-04-12 13:02:18 -07:00 |
|
|
Phil Wang
|
bd1e3b676e
|
get rid of numpy
0.8.1
|
2022-04-12 11:58:46 -07:00 |
|
|
Phil Wang
|
f4615599bc
|
use full attention at the center of the unet
0.8.0
|
2022-04-04 09:03:41 -07:00 |
|
|
Phil Wang
|
eb6e1b508e
|
greater kernel size in convnext blocks
0.7.1
|
2022-01-31 17:13:27 -08:00 |
|
|
Phil Wang
|
91cff45939
|
replace resnets with convnext blocks
0.7.0
|
2022-01-25 09:02:45 -08:00 |
|
|
Phil Wang
|
7b51e30da7
|
fix layernorm
0.6.9
|
2021-08-24 14:28:15 -07:00 |
|
|
Phil Wang
|
dadbf20154
|
remove stray print
0.6.8
|
2021-07-16 15:18:20 -07:00 |
|
|
Phil Wang
|
7706bdfc6f
|
use pre-layernorm with linear attention, and also allow for turning off time embedding
0.6.7
|
2021-06-25 10:48:11 -07:00 |
|
|
Phil Wang
|
183e5f3cc5
|
move all constants into configurable class init parameters
0.6.6
|
2021-06-25 10:37:36 -07:00 |
|
|
Phil Wang
|
16c9ae7bb3
|
fix data not being normalized to range of -1 to 1
0.6.5
|
2021-06-21 18:51:36 -07:00 |
|
|
Phil Wang
|
f5916111f8
|
0.6.3
0.6.3
|
2021-06-21 17:41:16 -07:00 |
|
|
Phil Wang
|
ad9e303ff3
|
fix channels
0.6.1
|
2021-06-11 15:28:36 -07:00 |
|
|
Phil Wang
|
ae42f48f6a
|
prepare so that unet can work with a channel of one, and also make it so image size is hard coded in diffusion class. preparing for training on protein distograms
0.6.0
|
2021-06-11 14:06:39 -07:00 |
|
|
Phil Wang
|
5989f4c77e
|
recommit sample
|
2020-10-13 09:32:56 -07:00 |
|
|
Phil Wang
|
2082046888
|
set higher num train steps, so non-practitioners do not think it is completed
|
2020-10-11 13:49:49 -07:00 |
|
|
Phil Wang
|
3c5b7e2d56
|
update readme
|
2020-10-10 10:37:44 -07:00 |
|
|
Phil Wang
|
d4ce9f6c38
|
save samples and models to ./results path
0.5.2
|
2020-10-09 21:50:23 -07:00 |
|
|
Phil Wang
|
ff451f697e
|
update with new and improved cosine noise scheduler
|
2020-10-09 21:21:02 -07:00 |
|
|
Phil Wang
|
3d96532c60
|
update citations in preparation to add improvements from a iclr 2021 paper
|
2020-10-05 17:15:28 -07:00 |
|