f1808e0970
Post crossvit merge cleanup, change model names to reflect input size, cleanup img size vs scale handling, fix tests
2021-09-13 11:49:54 -07:00
4027412757
Add resnet33ts weights, update resnext26ts baseline weights
2021-09-09 14:46:41 -07:00
9fe5798bee
fix bug for reset classifier and fix for validating the dimension
2021-09-08 21:58:17 -04:00
3718c5a5bd
fix loading pretrained model
2021-09-08 11:53:05 -04:00
bb50b69a57
fix for torch script
2021-09-08 11:20:59 -04:00
5bd04714e4
Cleanup weight init for byob/byoanet and related
2021-09-05 15:34:05 -07:00
8642401e88
Swap botnet 26/50 weights/models after realizing a mistake in arch def, now figuring out why they were so low...
2021-09-05 15:17:19 -07:00
5f12de4875
Add initial AttentionPool2d that's being trialed. Fix comment and still trying to improve reliability of sgd test.
2021-09-05 12:41:14 -07:00
76881d207b
Add baseline resnet26t @ 256x256 weights. Add 33ts variant of halonet with at least one halo in stage 2,3,4
2021-09-04 14:52:54 -07:00
484e61648d
Adding the attn series weights, tweaking model names, comments...
2021-09-03 18:09:42 -07:00
fb94350896
Update training script and loader factory to allow use of scheduler updates, repeat augment, and bce loss
2021-09-01 17:46:40 -07:00
f262137ff2
Add RepeatAugSampler as per DeiT RASampler impl, showing promise for current (distributed) training experiments.
2021-09-01 17:40:53 -07:00
ba9c1108a1
Add a BCE loss impl that converts dense targets to sparse /w smoothing as an alternate to CE w/ smoothing. For training experiments.
2021-09-01 17:39:28 -07:00
29a37e23ee
LR scheduler update:
...
* add polynomial decay 'poly'
* cleanup cycle specific args for cosine, poly, and tanh sched, t_mul -> cycle_mul, decay -> cycle_decay, default cycle_limit to 1 in each opt
* add k-decay for cosine and poly sched as per https://arxiv.org/abs/2004.05909
* change default tanh ub/lb to push inflection to later epochs
2021-09-01 17:33:11 -07:00
492c0a4e20
Update HaloAttn comment
2021-09-01 17:14:31 -07:00
7ab9d4555c
add crossvit
2021-09-01 17:13:12 -04:00
3b9032ea48
Use Tensor.unfold().unfold() for HaloAttn, fast like as_strided but more clarity
2021-08-27 12:45:53 -07:00
78933122c9
Fix silly typo
2021-08-27 09:22:20 -07:00
2568ffc5ef
Merge branch 'master' into attn_update
2021-08-27 09:21:22 -07:00
708d87a813
Fix ViT SAM weight compat as weights at URL changed to not use repr layer. Fix #825 . Tweak optim test.
2021-08-27 09:20:13 -07:00
8449ba210c
Improve performance of HaloAttn, change default dim calc. Some cleanup / fixes for byoanet. Rename resnet26ts to tfs to distinguish (extra fc).
2021-08-26 21:56:44 -07:00
a8b65695f1
Add resnet26ts and resnext26ts models for non-attn baselines
2021-08-21 12:42:10 -07:00
a5a542f17d
Fix typo
2021-08-20 17:47:23 -07:00
925e102982
Update attention / self-attn based models from a series of experiments:
...
* remove dud attention, involution + my swin attention adaptation don't seem worth keeping
* add or update several new 26/50 layer ResNe(X)t variants that were used in experiments
* remove models associated with dead-end or uninteresting experiment results
* weights coming soon...
2021-08-20 16:13:11 -07:00
d667351eac
Tweak accuracy topk safety. Fix #807
2021-08-19 14:18:53 -07:00
35c9740826
Fix accuracy when topk > num_classes
2021-08-19 11:58:59 +02:00
a16a753852
Add lamb/lars to optim init imports, remove stray comment
2021-08-18 22:55:02 -07:00
c207e02782
MOAR optimizer changes. Woo!
2021-08-18 22:20:35 -07:00
a426511c95
More optimizer cleanup. Change all to no longer use .data. Improve (b)float16 use with adabelief. Add XLA compatible Lars.
2021-08-18 17:21:56 -07:00
9541f4963b
One more scalar -> tensor fix for lamb optimizer
2021-08-18 11:20:25 -07:00
8f68193c91
Update lamp.py comment
2021-08-18 09:27:40 -07:00
4d284017b8
Merge pull request #813 from rwightman/opt_cleanup
...
Optimizer cleanup and additions
2021-08-18 09:12:00 -07:00
a6af48be64
add madgradw optimizer
2021-08-17 22:19:27 -07:00
55fb5eedf6
Remove experiment from lamb impl
2021-08-17 21:48:26 -07:00
8a9eca5157
A few optimizer comments, dead import, missing import
2021-08-17 18:01:33 -07:00
ac469b50da
Optimizer improvements, additions, cleanup
...
* Add MADGRAD code
* Fix Lamb (non-fused variant) to work w/ PyTorch XLA
* Tweak optimizer factory args (lr/learning_rate and opt/optimizer_name), may break compat
* Use newer fn signatures for all add,addcdiv, addcmul in optimizers
* Use upcoming PyTorch native Nadam if it's available
* Cleanup lookahead opt
* Add optimizer tests
* Remove novograd.py impl as it was messy, keep nvnovograd
* Make AdamP/SGDP work in channels_last layout
* Add rectified adablief mode (radabelief)
* Support a few more PyTorch optim, adamax, adagrad
2021-08-17 17:51:20 -07:00
abf3e044bb
Update scheduler_factory.py
...
remove duplicate code from create_scheduler()
2021-08-14 22:53:17 +02:00
3cdaf5ed56
Add `mmax` config key to auto_augment for increasing upper bound of RandAugment magnitude beyond 10. Make AugMix uniform sampling default not override config setting.
2021-08-12 15:39:05 -07:00
1042b8a146
Add non fused LAMB optimizer option
2021-08-09 13:13:43 -07:00
01cb46a9a5
Add gc_efficientnetv2_rw_t weights (global context instead of SE attn). Add TF XL weights even though the fine-tuned ones don't validate that well. Change default arg for GlobalContext to use scal (mul) mode.
2021-08-07 16:45:29 -07:00
d3f7440650
Add EfficientNetV2 XL model defs
2021-07-22 13:15:24 -07:00
72b227dcf5
Merge pull request #750 from drjinying/master
...
Specify "interpolation" mode in vision_transformer's resize_pos_embed
2021-07-13 11:01:20 -07:00
2907c1f967
Merge pull request #746 from samarth4149/master
...
Adding a Multi Step LR Scheduler
2021-07-13 10:55:54 -07:00
748ab852ca
Allow act_layer switch for xcit, fix in_chans for some variants
2021-07-12 13:27:29 -07:00
20b2d4b69d
Use bicubic interpolation in resize_pos_embed()
2021-07-12 10:38:31 -07:00
d3255adf8e
Merge branch 'xcit' of https://github.com/alexander-soare/pytorch-image-models into alexander-soare-xcit
2021-07-12 08:30:30 -07:00
f8039c7492
Fix gc effv2 model cfg name
2021-07-11 12:14:31 -07:00
3a55a30ed1
add notes from author
2021-07-11 14:25:58 +01:00
899cf84ccc
bug fix - missing _dist postfix for many of the 224_dist models
2021-07-11 12:41:51 +01:00
623e8b8eb8
wip xcit
2021-07-11 09:39:38 +01:00
392368e210
Add efficientnetv2_rw_t defs w/ weights, and gc variant, as well as gcresnet26ts for experiments. Version 0.4.13
2021-07-09 16:46:52 -07:00
daab57a6d9
1. Added a simple multi step LR scheduler
2021-07-09 16:18:27 -04:00
6d8272e92c
Add SAM pretrained model defs/weights for ViT B16 and B32 models.
2021-07-08 11:51:12 -07:00
ee4d8fc69a
Remove unecessary line from nest post refactor
2021-07-05 21:22:46 -07:00
8165cacd82
Realized LayerNorm2d won't work in all cases as is, fixed.
2021-07-05 18:21:34 -07:00
81cd6863c8
Move aggregation (convpool) for nest into NestLevel, cleanup and enable features_only use. Finalize weight url.
2021-07-05 18:20:49 -07:00
6ae0ac6420
Merge branch 'nested_transformer' of https://github.com/alexander-soare/pytorch-image-models into alexander-soare-nested_transformer
2021-07-03 12:45:26 -07:00
7b8a0017f1
wip to review
2021-07-03 12:10:12 +01:00
b11d949a06
wip checkpoint with some feature extraction work
2021-07-03 11:45:19 +01:00
23bb72ce5e
nested_transformer wip
2021-07-02 20:12:29 +01:00
766b4d3262
Fix features for resnetv2_50t
2021-06-28 15:56:24 -07:00
e8045e712f
Fix BatchNorm for ResNetV2 non GN models, add more ResNetV2 model defs for future experimentation, fix zero_init of last residual for pre-act.
2021-06-28 10:52:45 -07:00
20a2be14c3
Add gMLP-S weights, 79.6 top-1
2021-06-23 10:40:30 -07:00
85f894e03d
Fix ViT in21k representation (pre_logits) layer handling across old and new npz checkpoints
2021-06-23 10:38:34 -07:00
b41cffaa93
Fix a few issues loading pretrained vit/bit npz weights w/ num_classes=0 __init__ arg. Missed a few other small classifier handling detail on Mlp, GhostNet, Levit. Should fix #713
2021-06-22 23:16:05 -07:00
9c9755a808
AugReg release
2021-06-20 17:46:06 -07:00
381b279785
Add hybrid model fwds back
2021-06-19 22:28:44 -07:00
26f04a8e3e
Fix a weight link
2021-06-19 16:39:36 -07:00
8f4a0222ed
Add GMixer-24 MLP model weights, trained w/ TPU + PyTorch XLA
2021-06-18 16:49:28 -07:00
4c09a2f169
Bump version 0.4.12
2021-06-18 16:17:34 -07:00
b319eb5b5d
Update ViT weights, more details to be added before merge.
2021-06-18 16:16:49 -07:00
8257b86550
Fix up resnetv2 bit/bitm model default res
2021-06-18 16:16:06 -07:00
1228f5a3d8
Add BiT distilled 50x1 and teacher 152x2 models from 'A good teacher is patient and consistent' paper.
2021-06-18 11:40:33 -07:00
511a8e8c96
Add official ResMLP weights.
2021-06-14 17:03:16 -07:00
b9cfb64412
Support npz custom load for vision transformer hybrid models. Add posembed rescale for npz load.
2021-06-14 12:31:44 -07:00
8319e0c373
Add file docstring to std_conv.py
2021-06-13 12:31:06 -07:00
4d96165989
Merge branch 'master' into cleanup_xla_model_fixes
2021-06-12 23:19:25 -07:00
8880f696b6
Refactoring, cleanup, improved test coverage.
...
* Add eca_nfnet_l2 weights, 84.7 @ 384x384
* All 'non-std' (ie transformer / mlp) models have classifier / default_cfg test added
* Fix #694 reset_classifer / num_features / forward_features / num_classes=0 consistency for transformer / mlp models
* Add direct loading of npz to vision transformer (pure transformer so far, hybrid to come)
* Rename vit_deit* to deit_*
* Remove some deprecated vit hybrid model defs
* Clean up classifier flatten for conv classifiers and unusual cases (mobilenetv3/ghostnet)
* Remove explicit model fns for levit conv, just pass in arg
2021-06-12 16:40:02 -07:00
ba2ca4b464
One codepath for stdconv, switch layernorm to batchnorm so gain included. Tweak epsilon values for nfnet, resnetv2, vit hybrid.
2021-06-12 12:27:43 -07:00
b7a568f065
Fix torchscript issue in bat
2021-06-08 23:19:51 -07:00
d17b374f0f
Minimum input_size needed to be higher
2021-06-08 21:31:39 -07:00
b3b90d944d
Add min_input_size to bat_resnext to prevent test breakage.
2021-06-08 17:32:08 -07:00
d413eef1bf
Add ResMLP-24 model weights that I trained in PyTorch XLA on TPU-VM. 79.2 top-1.
2021-06-08 14:22:05 -07:00
10d8fa4620
Add gc and bat attention resnext26ts variants to byob for test.
2021-06-08 14:21:07 -07:00
2f5ed2dec1
Update `init_values` const for 24 and 36 layer ResMLP models
2021-06-07 17:15:04 -07:00
8e4ac3549f
All ScaledStdConv and StdConv uses default to using F.layernorm so that they work with PyTorch XLA. eps value tweaking is a WIP.
2021-06-07 17:14:19 -07:00
2a63d0246b
Post merge cleanup
2021-06-07 14:38:30 -07:00
45dec179e5
Merge pull request #681 from lmk123568/master
...
Update convit.py
2021-06-07 14:10:53 -07:00
ded1671483
Fix stochastic depth working only with a shortcut
2021-06-07 23:08:55 +09:00
b87d98b238
Update convit.py
...
Cut out the duplicates
2021-06-06 17:58:31 +08:00
02320c3e3d
Bump version to 0.4.11
2021-05-31 15:41:51 -07:00
bda8ab015a
Remove min channels for SelectiveKernel, divisor should cover cases well enough.
2021-05-31 15:38:56 -07:00
a27f4aec4a
Missed args for skresnext w/ refactoring.
2021-05-31 14:06:34 -07:00
307a935b79
Add non-local and BAT attention. Merge attn and self-attn factories into one. Add attention references to README. Add mlp 'mode' to ECA.
2021-05-31 13:18:11 -07:00
8bf63b6c6c
Able to use other attn layer in EfficientNet now. Create test ECA + GC B0 configs. Make ECA more configurable.
2021-05-30 12:47:02 -07:00
bcec14d3b5
Bring EfficientNet SE layer in line with others, pull se_ratio outside of blocks. Allows swapping w/ other attn layers.
2021-05-29 23:41:38 -07:00
9611458e19
Throw in some FBNetV3 code I had lying around, some refactoring of SE reduction channel calcs for all EffNet archs.
2021-05-28 20:47:24 -07:00
01b9108619
Merge branch 'master' into more_attn
2021-05-28 11:09:37 -07:00
d7bab8a6c5
Fix strict flag change for checkpoint load.
2021-05-28 09:54:50 -07:00
02f9d4bc34
Add weights for resnet51q model, add 61q def.
2021-05-28 09:53:16 -07:00
Ross Wightman
Richard Chen
Yohann Lereclus
Ross Wightman
Sepehr Sameni
Ying Jin
Alexander Soare
samarth
Dongyoon Han
Mike