79abc25f55
Add ADOPT optimizer
2024-11-12 20:49:01 -08:00
36a45e5d94
Improve row/col dim var name
2024-11-12 20:49:01 -08:00
e7b0480381
Cleanup original adafactor impl, add row/col dim heuristic that works with both conv and linear layers
2024-11-12 20:49:01 -08:00
1409ce2dbe
Change eps defaults in adafactor_bv again after some checking
2024-11-12 20:49:01 -08:00
9d8ccd2ba7
A bit of lars/lamb cleanup, torch.where supports scalars properly now, make lamb grad clipping optional, clean it up a bit
2024-11-12 20:49:01 -08:00
7cfaeced67
Change adafactor_bv epsilon default
2024-11-12 20:49:01 -08:00
0b5ae49251
Remove adafactorbv numpy dep, hack fix for loading optimizer state w/ half prec momentum (need better one)
2024-11-12 20:49:01 -08:00
19090ea966
Need to init momentum with correct dtype
2024-11-12 20:49:01 -08:00
484a88f4b4
Remove unused beta2 fn, make eps grad^2 handling same across factorized and non-factorized cases
2024-11-12 20:49:01 -08:00
7c16adca83
An impl of adafactor as per big vision (scaling vit) changes
2024-11-12 20:49:01 -08:00
363b043c13
Extend train epoch schedule by warmup_epochs if warmup_prefix enable, allows schedule to reach end w/ prefix enabledy
2024-11-08 11:01:11 -08:00
7f0c1b1f30
Add trust_remote_code argument to ReaderHfds
2024-11-08 08:16:36 -08:00
eb94efb218
fix pos embed dynamic resampling for eva
2024-11-06 16:03:27 -08:00
3c7822c621
fix pos embed dynamic resampling for deit
2024-11-06 16:03:27 -08:00
3ae3f44288
Fix positional embedding resampling for non-square inputs in ViT
2024-11-06 16:03:27 -08:00
d4dde48dd5
Missed first_conv from resnet18d
2024-10-31 19:29:53 -07:00
e6263bf64d
Add resnet and resnet-v2 18/34 weights trained with mnv4 small based recipe
2024-10-31 16:39:35 -07:00
f5b58e31a2
Allow non train mode for wds reader to operate w/o sample count, exhaust iterator
2024-10-31 16:39:35 -07:00
f689c850b9
One more small c&p issue
2024-10-23 21:51:09 -07:00
baa7242dd3
Fix c&p error, slight reformat
2024-10-23 21:51:09 -07:00
1b5cae681c
Update some clip pretrained weights to point to new hub locations, add a few missing weights
2024-10-23 21:51:09 -07:00
310ffa32c5
Update version.py
...
dev version 1.0.12.dev0
2024-10-19 09:56:17 -07:00
015fbe457a
Merge branch 'MengqingCao-npu_support' into device_amp_cleanup
2024-10-18 14:50:44 -07:00
81b59faf77
Merge branch 'npu_support' of github.com:MengqingCao/pytorch-image-models into MengqingCao-npu_support
2024-10-18 14:50:00 -07:00
1766a01f96
Cleanup some amp related behaviour to better support different (non-cuda) devices
2024-10-18 13:54:16 -07:00
37c731ca37
fix device check
2024-10-17 12:38:02 +00:00
ca20e102fe
mambaout.py: fixed bug
2024-10-17 01:03:28 +03:30
8cb2548962
Version 1.0.11
2024-10-16 14:14:44 -07:00
89dffc5ff0
Another small fix for original mambaout models, no classifier nn.Linear when num_classe=0 on init
2024-10-16 12:36:36 -07:00
fad4538801
Elevate import deprecation warnings from DeprecationWarning to FutureWarning so messages are now seen
2024-10-16 11:30:01 -07:00
a1f379e712
Add intern300m vit w/ converted timm weights. Fix #2300
2024-10-16 10:29:06 -07:00
234f975787
add npu support
2024-10-16 07:13:45 +00:00
60f517c883
Fix wrong name in _all_ for models._registry
2024-10-15 07:39:46 -07:00
b4a9a166c3
Version 1.0.10
2024-10-14 21:40:30 -07:00
c3052fa19e
Merge pull request #2298 from huggingface/preact_resnet18
...
Add resnet18/18d pre-act model configs for potential training.
2024-10-14 19:39:04 -07:00
abdf33145c
Add 34/34d pre-act resnet variants
2024-10-14 13:23:50 -07:00
c82ce86f8f
Add 384x384 mambaout_base_plus model weights
2024-10-14 12:28:57 -07:00
82ae247879
MambaOut weights on hub, configs finalized
2024-10-11 11:07:40 -07:00
7efb60c299
Add first_conv for mambaout
2024-10-09 14:11:40 -07:00
5dc5ee5b42
Add global_pool to mambaout __init__ and pass to heads
2024-10-09 14:11:40 -07:00
9d1dfe8dbe
Incorrectly named head_hidden_size
2024-10-09 14:11:40 -07:00
91e743f2dd
Mambaout tweaks
2024-10-09 14:11:40 -07:00
4542cf03f9
Add features_only, other bits to mambaout, define different base alternatives
2024-10-09 14:11:40 -07:00
c2da12c7e1
Update rw models, fix heads
2024-10-09 14:11:40 -07:00
f2086f51a0
Add mambaout builder support, pretrained weight remap
2024-10-09 14:11:40 -07:00
c6ef54eefa
Initial mambaout work
2024-10-09 14:11:40 -07:00
d9321b0e10
Add weights for fine-tuned siglip so400m. Add webli_i18n pretrained tags for the multi-lingual model variants (incl older base)
2024-10-09 09:04:44 -07:00
01b62264af
Add i18n variant of so400m model w/ weights. Add two in1k fine-tunes of original so400m 384x384 but at 378x378 (better matches patch14)
2024-10-08 23:40:24 -07:00
72f0edb7e8
missed first_conv for rnv2 18d
2024-10-08 12:38:54 -07:00
3ed603a2ce
Add resnet18/18d pre-act model configs for potential training. Fix #2289
2024-10-08 11:28:07 -07:00
41a79e0fcb
Add overlapped stem convnext zepto weights
2024-10-08 11:26:34 -07:00
545bd4056c
Tag along test_vit3 weights
2024-09-30 12:03:32 -07:00
69b687d4cc
Add zepto weights
2024-09-30 11:43:23 -07:00
c6e5557a5a
Mismatch pretrained_cfg
2024-09-30 11:43:23 -07:00
5d7bd2973e
convnext zepto, rmsnorm experiments
2024-09-30 11:43:23 -07:00
e3242a5258
Merge pull request #2277 from huggingface/more_tiny_test_models
...
Adding some more tiny test models to train...
2024-09-22 10:28:29 -07:00
c1cb5641c7
Add weight fore mobilenetv4 small 0.5, change 0.25 -> 0.35
2024-09-22 10:27:01 -07:00
a22ce0a329
Merge branch 'patch-1' of https://github.com/baorepo/pytorch-image-models into baorepo-patch-1
2024-09-22 10:14:35 -07:00
9067be6a30
Add weights for new tiny test models
2024-09-22 07:59:23 -07:00
65564f7da5
Fix reversed H & W padding for swin patch merging
2024-09-21 16:51:02 -07:00
a2f539f055
Add a few more test model defs in prep for weight upload
2024-09-21 11:38:38 -07:00
6ab2af610d
Adding some more tiny test models to train
2024-09-06 15:35:57 -07:00
d6b8816eda
MobilenetV4: add two more lightweight models
...
Mobilenetv4 is very fast and ideal for embedded devices. However, for many low-cost, low-power embedded MCU devices, smaller models are required. Hopefully this PR will merge.
2024-09-05 02:34:11 +00:00
f81cbdcca9
Merge pull request #2274 from huggingface/bulk_runner_tweaks
...
Better all res resolution for bulk runner
2024-09-03 12:11:56 -07:00
6c42299b3e
Back to dev version
2024-09-03 10:29:05 -07:00
a50713ce6e
Fix #2272
2024-09-02 13:20:05 -07:00
ebbe530ee4
Add MobileNetV3 RA4 (mnv4 recipe) weights
2024-09-02 13:10:34 -07:00
fa4a1e597f
Better all res resolution for bulk runner
2024-08-26 22:28:01 -07:00
072783331f
Version 1.0.9
2024-08-23 11:47:42 -07:00
76b0e9931a
Placeholder for new mnv3 model
2024-08-23 10:11:20 -07:00
39e92f0c0d
mobilenet_edgetpu can use group_size override, more consistency in arg wrap/sadface w/ extra group_size arg
2024-08-22 11:44:02 -07:00
b9f020a509
Allow group_size override for more efficientnet and mobilenetv3 based models
2024-08-21 16:51:38 -07:00
17923a66bb
Add layer scale to hieradet
2024-08-21 11:23:39 -07:00
47e6958263
Add hierdet_small (non sam) model def
2024-08-21 11:05:54 -07:00
9fcbf39cdc
Add remaining sbb vit betwixt/mediumd fine-tunes
2024-08-21 10:09:38 -07:00
dc94cca0e5
Remaining Hiera sbb weights uploaded
2024-08-21 10:06:27 -07:00
a256e50457
Move padding back in front of windowing
2024-08-17 11:22:53 -07:00
7d83749207
pool size test fixes
2024-08-17 08:27:13 -07:00
1bd92bca0e
Add fused_attn flag to HieraDet attn block
2024-08-16 22:57:49 -07:00
691bb54443
Larger min input size needed
2024-08-16 17:09:19 -07:00
de3a91a7a0
Add min_input_size of 128 for hieradet/sam2
2024-08-16 15:13:56 -07:00
0b05122cda
Fixing hieradet (sam2) tests
2024-08-16 14:33:40 -07:00
e035381171
Move padding out of windowing code for hieradet, fix torchscript typing issues, make pooling MaxPool unique instances across two modules
2024-08-16 13:36:33 -07:00
146c2fbe34
Add resnet50d and efficientnet_b1 ra4 (mnv4) hparam weights
2024-08-16 12:10:00 -07:00
962958723c
More Hiera updates. Add forward_intermediates to hieradat/sam2 impl. Make both use same classifier module. Add coarse bool to intermediates.
2024-08-16 11:10:04 -07:00
f2cfb4c677
Add WIP HieraDet impl (SAM2 backbone support)
2024-08-15 17:58:15 -07:00
a50e53d41f
Rename global pos embed for Hiera abswin, factor out commonly used vit weight init fns to layers. Add a channels-last ver of normmlp head.
2024-08-15 17:46:36 -07:00
2f3fed43b8
Fix hiera init with num_classes=0, fix weight tag names for sbb2 hiera/vit weights, add LayerScale/LayerScale2d to layers
2024-08-15 11:14:38 -07:00
fee91fdd41
Update Hiera model for abswin, more stable weight init, layer-scale. ImageNet-12k weights for hiera_small_abswin, and two of the sbb vits with improved reg4 init.
2024-08-14 12:22:40 -07:00
ac3470188b
Add mobilenetv1_125 weights from ra4 / mnv4 hparam series
2024-08-14 08:47:02 -07:00
a7b0bfc376
exception for unknown arg should have been TypeError
2024-08-12 16:09:36 -07:00
bd0f79153b
Add weights only flag to avoid warning, try to keep bwd compat. Default to True for remote load of pretrained weights, keep False for local checkpoing load to avoid training checkpoint breaks.. fix #2249
2024-08-12 11:41:41 -07:00
531215eded
Bring changelog up to current, back to dev version, 1.0.9.dev0
2024-08-08 09:17:23 -07:00
bf781bc3c2
Merge pull request #2245 from dhkim0225/feature/rdnet
...
Feature: Add RDNet
2024-08-07 16:19:22 -07:00
49f6e1da96
Remove eval to avoid safety concerns
2024-08-07 14:54:13 -07:00
2d5c9bf60d
Merge pull request #2253 from brendanartley/hgnet-grad-checkpointing
...
Add gradient checkpointing to hgnets
2024-08-07 12:45:14 -07:00
e9ef9424f0
Add a few missing __all__ entries.
2024-08-07 09:35:51 -07:00
232a459e36
Added gradient checkpointing to hgnet
2024-08-07 09:22:51 -07:00
ea33d6a74d
bugfix pool_size
2024-08-03 09:50:20 +09:00
10344625be
proj_drop wasn't being called in sam vit
2024-08-02 16:12:02 -07:00
6fdc9d8243
update code for ci pass
2024-08-02 15:14:40 +09:00
7a866b6521
update code for torchscript
2024-08-02 09:58:13 +09:00
8248122f82
add rdnet
2024-08-01 14:54:29 +09:00
025259024d
add rdnet
2024-08-01 14:51:15 +09:00
225f4f92b3
add rdnet
2024-08-01 14:49:21 +09:00
4a10302754
Add mobilenet_edgetpu_v2_m weights
2024-07-28 17:19:36 -07:00
70182ea0cd
Version 1.0.8
2024-07-26 17:05:36 -07:00
ab8cb070fc
Add xavier_uniform init of MNVC hybrid attention modules. Small improvement in training stability.
2024-07-26 17:03:40 -07:00
cec70b6779
Merge pull request #2225 from huggingface/small_things
...
Small things
2024-07-25 20:29:13 -07:00
61df3fde89
Wrong hybrid_medium in12k pool sizes
2024-07-25 15:39:21 -07:00
9aa2930760
Add latest mobilenetv4 and baseline updates for mobilenetv1 and efficientnet_b0 weights
2024-07-25 14:20:54 -07:00
7b6a406474
remove swin debug prints
2024-07-24 21:05:56 -07:00
4c531be479
set_input_size(), always_partition, strict_img_size, dynamic mask option for all swin models. More flexibility in resolution, window resizing.
2024-07-24 16:41:31 -07:00
8efdc38213
Fix #2242 add checks for out indices with intermediate getter mode
2024-07-23 08:19:09 -07:00
d2240745d3
Fix issue where feature out_indices out of order after wrapping with FeatureGetterNet due to use of set()
2024-07-22 13:33:30 -07:00
2b3f1a4633
Make channels for classic resnet configurable
2024-07-22 10:47:40 -07:00
9b2b8014e8
Add weights for test models
2024-07-22 10:08:57 -07:00
7e0caa1ba3
Padding helpers work if tuples/lists passed
2024-07-19 14:28:03 -07:00
2180800646
MQA query_strides bugs fix #2237 . No padding for avg_pool2d if not 'same', use scale_factor for Upsample.
2024-07-19 14:26:54 -07:00
1a05ed29a1
Add to 'abswin' hiera models for train trials
2024-07-19 11:05:31 -07:00
0cbf4fa586
_orig_mod still causing issues even though I thought it was fixed in pytorch, add unwrap / clean helpers
2024-07-19 11:03:45 -07:00
4cca568bd8
eva.py: fixed bug in applying attention mask
...
The mask should be applied before the softmax.
2024-07-19 15:12:04 +03:30
3a8a965891
Implement absolute+window pos embed for hiera, resizable but needs new weights
2024-07-18 21:43:37 -07:00
7160af4a24
Merge pull request #2229 from Promisery/reg_token
...
Initialize weights of reg_token for ViT
2024-07-18 09:25:29 -07:00
392b78aee7
set_input_size initial impl for vit & swin v1. Move HybridEmbed to own location in timm/layers
2024-07-17 15:25:48 -07:00
34c9fee554
Fix pass through of input / target keys so ImageDataset readers so args work with hfds instead of just hfids (iterable)
2024-07-17 10:11:46 -07:00
417cf7f871
Initialize weights of reg_token for ViT
2024-07-13 11:11:42 +08:00
f920119f3b
Fixing tests
2024-07-09 14:53:20 -07:00
644abf9588
Fix default_cfg test for mobilenet_100
2024-07-09 12:52:24 -07:00
d5afe106dc
Merge remote-tracking branch 'origin/tiny_test_models' into small_things
2024-07-09 12:49:57 -07:00
55101028bb
Rename test_tiny* -> test*. Fix ByobNet BasicBlock attn location and add test_byobnet model.
2024-07-09 11:53:11 -07:00
1334598462
Add support back to EfficientNet to disable head_conv / bn2 so mobilnetv1 can be implemented properly
2024-07-08 13:51:26 -07:00
800405d941
Add conv_large mobilenetv3 aa/blur model defs
2024-07-08 13:50:05 -07:00
f81b094aaa
Add 'qkv_bias_separate' flag for EVA/beit/swinv2 attn modules to allow an override for easy quantization wrappers. Fix #2098
2024-07-08 13:48:38 -07:00
83c2c2f0c5
Add 'Maybe' PIL / image tensor conversions in case image alread in tensor format
2024-07-08 13:43:51 -07:00
c01a47c9e7
Fix typo in type annotations in timm.models.hrnet
2024-07-08 00:53:16 +02:00
197c10463b
Fix jit.script breaking with features_fx
2024-06-28 03:58:51 +00:00
b751da692d
Add latest ix (xavier init for mqa) hybrid medium & large weights for MobileNetV4
2024-06-24 13:49:55 -07:00
d4d4d84fda
Dev version 1.0.8.dev0
2024-06-24 11:34:13 -07:00
f8342a045a
Merge pull request #2213 from huggingface/florence2
...
Fix #2212 map florence2 image tower to davit with a few changes
2024-06-24 11:01:08 -07:00
c33a001397
Fix typo
2024-06-24 21:54:38 +09:00
02d0f27721
cleanup davit padding
2024-06-22 12:06:46 -07:00
c715c724e7
Fix tracing by removing float cast, should end up float anyways
2024-06-22 08:35:30 -07:00
fb58a73033
Fix #2212 map florence2 image tower to davit with a few changes
2024-06-21 15:31:29 -07:00
b28945ff05
Version 1.0.7, prep for release
2024-06-18 16:19:43 -07:00
fb13e6385e
Merge pull request #2203 from huggingface/more_mobile
...
Add mobilenet edgetpu defs for exp, add ol mobilenet v1 back for comp…
2024-06-18 15:20:01 -07:00
16e082e1c2
Add mobilenetv4 hybrid-large weights
2024-06-17 11:08:31 -07:00
e41125cc83
Merge pull request #2209 from huggingface/fcossio-vit-maxpool
...
ViT pooling refactor
2024-06-17 07:51:12 -07:00
a22466852d
Add 2400 epoch mobilenetv4 small weights, almost at paper, rounds to 73.8
2024-06-16 10:51:00 -07:00
b1a6f4a946
Some missed reset_classifier() type annotations
2024-06-16 10:39:27 -07:00
Ross Wightman
Augustin Godinot
Wojtek Jasiński
Ross Wightman
MengqingCao
Feraidoon Mehri
alias pillar1989
brendanartley
dong-hyun
Promisery
Steffen Schneider
Daniel Suess
Sejik