Commit Graph

148 Commits

Author SHA1 Message Date
lucidrains
9a8e509b27 separate a simple vit from mp3, so that simple vit can be used after being pretrained 2023-03-07 19:31:10 -08:00
Phil Wang
258dd8c7c6 release mp3, contributed by @Vishu26 2023-03-07 14:29:45 -08:00
Phil Wang
5699ed7d13 double down on dual patch norm, fix MAE and Simmim to be compatible with dual patchnorm 2023-02-10 10:39:50 -08:00
Phil Wang
46dcaf23d8 seeing a signal with dual patchnorm in another repository, fully incorporate 2023-02-06 09:45:12 -08:00
Phil Wang
bdaf2d1491 adopt dual patchnorm paper for as many vit as applicable, release 1.0.0 2023-02-03 08:11:29 -08:00
Phil Wang
500e23105a need simple vit with patch dropout for another project 2022-12-05 10:47:36 -08:00
Phil Wang
89e1996c8b add vit with patch dropout, fully embrace structured dropout as multiple papers are now corroborating each other 2022-12-02 11:28:11 -08:00
Phil Wang
2f87c0cf8f offer 1d versions, in light of https://arxiv.org/abs/2211.14730 2022-12-01 10:31:05 -08:00
Phil Wang
59c8948c6a try to fix tests 2022-10-29 11:44:17 -07:00
Phil Wang
cb6d749821 add a 3d version of cct, addressing https://github.com/lucidrains/vit-pytorch/issues/238 0.38.1 2022-10-29 11:35:06 -07:00
Phil Wang
6ec8fdaa6d make sure global average pool can be used for vivit in place of cls token 2022-10-24 19:59:48 -07:00
Phil Wang
13fabf901e add vivit 2022-10-24 09:34:04 -07:00
Phil Wang
5f1a6a05e9 release updated mae where one can more easily visualize reconstructions, thanks to @Vishu26 2022-10-17 10:41:46 -07:00
Phil Wang
b4853d39c2 add the 3d simple vit 2022-10-16 20:45:30 -07:00
Phil Wang
29fbf0aff4 begin extending some of the architectures over to 3d, starting with basic ViT 2022-10-16 15:31:59 -07:00
Phil Wang
f86e052c05 offer way for extractor to return latents without detaching them 2022-07-16 16:22:40 -07:00
Phil Wang
2fa2b62def slightly more clear of einops rearrange for cls token, for https://github.com/lucidrains/vit-pytorch/issues/224 2022-06-30 08:11:17 -07:00
Phil Wang
9f87d1c43b follow @arquolo feedback and advice for MaxViT 2022-06-29 08:53:09 -07:00
Phil Wang
2c6dd7010a fix hidden dimension in MaxViT thanks to @arquolo 2022-06-24 23:28:35 -07:00
Phil Wang
6460119f65 be able to accept a reference to a layer within the model for forward hooking and extracting the embedding output, for regionvit to work with extractor 2022-06-19 08:22:18 -07:00
Phil Wang
4e62e5f05e make extractor flexible for layers that output multiple tensors, show CrossViT example 2022-06-19 08:11:41 -07:00
Phil Wang
b3e90a2652 add simple vit, from https://arxiv.org/abs/2205.01580 2022-05-03 20:24:14 -07:00
Phil Wang
4ef72fc4dc add EsViT, by popular request, an alternative to Dino that is compatible with efficient ViTs with accounting for regional self-supervised loss 2022-05-03 10:29:29 -07:00
Phil Wang
81661e3966 fix mbconv residual block 2022-04-06 16:43:06 -07:00
Phil Wang
13f8e123bb fix maxvit - need feedforwards after attention 2022-04-06 16:34:40 -07:00
Phil Wang
c7bb5fc43f maxvit intent to build (#211)
complete hybrid mbconv + block / grid efficient self attention MaxViT
2022-04-06 16:12:17 -07:00
Phil Wang
d93cd84ccd let windowed tokens exchange information across heads a la talking heads prior to pointwise attention in sep-vit 2022-03-31 15:22:24 -07:00
Phil Wang
5d4c798949 cleanup sepvit 2022-03-31 14:35:11 -07:00
Phil Wang
d65a742efe intent to build (#210)
complete SepViT, from bytedance AI labs
2022-03-31 14:30:23 -07:00
Phil Wang
8c54e01492 do not layernorm on last transformer block for scalable vit, as there is already one in mlp head 2022-03-31 13:25:21 -07:00
Phil Wang
df656fe7c7 complete learnable memory ViT, for efficient fine-tuning and potentially plays into continual learning 2022-03-31 09:51:12 -07:00
Phil Wang
4e6a42a0ca correct need for post-attention dropout 2022-03-30 10:50:57 -07:00
Phil Wang
9cd56ff29b CCT allow for rectangular images 2022-03-26 14:02:49 -07:00
Phil Wang
2aae406ce8 add proposed parallel vit from facebook ai for exploration purposes 2022-03-23 10:42:35 -07:00
Phil Wang
c2b2db2a54 fix window size of none for scalable vit for rectangular images 2022-03-22 17:37:59 -07:00
Phil Wang
719048d1bd some better defaults for scalable vit 2022-03-22 17:19:58 -07:00
Phil Wang
d27721a85a add scalable vit, from bytedance AI 2022-03-22 17:02:47 -07:00
Phil Wang
cb22cbbd19 update to einops 0.4, which is torchscript jit friendly 2022-03-22 13:58:00 -07:00
Phil Wang
6db20debb4 add patch merger 2022-03-01 16:50:17 -08:00
Phil Wang
1bae5d3cc5 allow for rectangular images for efficient adapter 2022-01-31 08:55:31 -08:00
Phil Wang
25b384297d return None from extractor if no attention layers 2022-01-28 17:49:58 -08:00
Phil Wang
64a07f50e6 epsilon should be inside square root 2022-01-24 17:24:41 -08:00
Phil Wang
c1528acd46 fix feature maps in Nest, thanks to @MarkYangjiayi 2022-01-22 13:17:30 -08:00
Phil Wang
1cc0f182a6 decoder positional embedding needs to be reapplied https://twitter.com/giffmana/status/1479195631587631104 2022-01-06 13:14:41 -08:00
Phil Wang
28eaba6115 0.26.2 2022-01-03 12:56:34 -08:00
Phil Wang
91ed738731 0.26.1 2021-12-30 19:31:26 -08:00
Phil Wang
70ba532599 add ViT for small datasets https://arxiv.org/abs/2112.13492 2021-12-28 10:58:21 -08:00
Phil Wang
e52ac41955 allow extractor to only return embeddings, to ready for vision transformers to be used in x-clip 2021-12-25 12:31:21 -08:00
Phil Wang
0891885485 include tests in package for conda 2021-12-22 12:44:29 -08:00
Phil Wang
976f489230 add some tests 2021-12-22 09:13:31 -08:00