Commit Graph

891 Commits (30bd1746c55974181ed2b24bb68c55ac60c68ce0)

Author SHA1 Message Date
Ross Wightman 4b30bae67b Add updated vit_relpos weights, and impl w/ support for official swin-v2 differences for relpos. Add bias control support for MLP layers
3 years ago
Ross Wightman d4c0588012 Remove persistent buffers from Swin-V2. Change SwinV2Cr cos attn + tau/logit_scale to match official, add ckpt convert, init_value zeros resid LN weight by default
3 years ago
Ross Wightman 27c42f0830 Fix torchscript use for offician Swin-V2, add support for non-square window/shift to WindowAttn/Block
3 years ago
Ross Wightman 2f2b22d8c7 Disable nvfuser fma / opt level overrides per #1244
3 years ago
Ross Wightman c0211b0bf7 Swin-V2 test fixes, typo
3 years ago
Ross Wightman 9a86b900fa Official SwinV2 models
3 years ago
Ross Wightman d07d015173
Merge pull request #1249 from okojoalg/sequencer
3 years ago
Ross Wightman d30685c283
Merge pull request #1251 from hankyul2/fix-multistep-scheduler
3 years ago
han a16171335b fix: change milestones to decay-milestones
3 years ago
Ross Wightman 39b725e1c9 Fix tests for rank-4 output where feature channels dim is -1 (3) and not 1
3 years ago
Ross Wightman 78a32655fa Fix poolformer group_matcher to merge proj downsample with previous block, support coarse
3 years ago
Ross Wightman d79f3d9d1e Fix torchscript use for sequencer, add group_matcher, forward_head support, minor formatting
3 years ago
Ross Wightman 37b6920df3 Fix group_matcher regex for regnet.py
3 years ago
okojoalg 93a79a3dd9 Fix num_features in Sequencer
3 years ago
han 57a988df30 fix: multistep lr decay epoch bugs
3 years ago
okojoalg 578d52e752 Add Sequencer
3 years ago
Ross Wightman f5ca4141f7 Adjust arg order for recent vit model args, add a few comments
3 years ago
Ross Wightman 41dc49a337 Vision Transformer refactoring and Rel Pos impl
3 years ago
Ross Wightman b7cb8d0337 Add Swin-V2 Small-NS weights (83.5 @ 224). Add layer scale like 'init_values' via post-norm LN weight scaling
3 years ago
jjsjann123 f88c606fcf fixing channels_last on cond_conv2d; update nvfuser debug env variable
3 years ago
Li Dong 09e9f3defb
migrate azure blob for beit checkpoints
3 years ago
Ross Wightman 52ac881402 Missed first_conv in latest seresnext 'D' default_cfgs
3 years ago
Ross Wightman 7629d8264d Add two new SE-ResNeXt101-D 32x8d weights, one anti-aliased and one not. Reshuffle default_cfgs vs model entrypoints for resnet.py so they are better aligned.
3 years ago
SeeFun 8f0bc0591e fix convnext args
3 years ago
Ross Wightman c5a8e929fb Add initial swinv2 tiny / small weights
3 years ago
Ross Wightman f670d98cb8 Make a few more layers symbolically traceable (remove from FX leaf modules)
3 years ago
SeeFun ec4e9aa5a0
Add ConvNeXt tiny and small pretrain in22k
3 years ago
Ross Wightman 575924ed60 Update test crop for new RegNet-V weights to match Y
3 years ago
Ross Wightman 1618527098 Add layer scale and parallel blocks to vision_transformer
3 years ago
Ross Wightman c42be74621 Add attrib / comments about Swin-S3 (AutoFormerV2) weights
3 years ago
Ross Wightman 474ac906a2 Add 'head norm first' convnext_tiny_hnf weights
3 years ago
Ross Wightman dc51334cdc Fix pruned adapt for EfficientNet models that are now using BatchNormAct layers
3 years ago
Ross Wightman 024fc4d9ab version 0.6.1 for master
3 years ago
Ross Wightman e1e037ba52 Fix bad tuple typing fix that was on XLA branch bust missed on master merge
3 years ago
Ross Wightman 341b464a5a Remove redundant noise attr from Plateau scheduler (use parent)
3 years ago
Ross Wightman fe457c1996 Update SwinTransformerV2Cr post-merge, update with grad checkpointing / grad matcher
3 years ago
Ross Wightman b049a5c5c6 Merge remote-tracking branch 'origin/master' into norm_norm_norm
3 years ago
Ross Wightman 7cdd164d77 Fix #1184, scheduler noise bug during merge madness
3 years ago
Ross Wightman 9440a50c95 Merge branch 'mrT23-master'
3 years ago
Ross Wightman d98aa47d12 Revert ml-decoder changes to model factory and train script
3 years ago
Ross Wightman b20665d379
Merge pull request #1007 from qwertyforce/patch-1
3 years ago
Ross Wightman 7a0994f581
Merge pull request #1150 from ChristophReich1996/master
3 years ago
Ross Wightman 61d3493f87 Fix hf-hub handling when hf-hub is config source
3 years ago
Ross Wightman 5f47518f27 Fix pit implementation to be clsoer to deit/levit re distillation head handling
3 years ago
Ross Wightman 0862e6ebae Fix correctness of some group matching regex (no impact on result), some formatting, missed forward_head for resnet
3 years ago
Ross Wightman 94bcdebd73 Add latest weights trained on TPU-v3 VM instances
3 years ago
Ross Wightman 0557c8257d Fix bug introduced in non layer_decay weight_decay application. Remove debug print, fix arg desc.
3 years ago
Ross Wightman 372ad5fa0d Significant model refactor and additions:
3 years ago
Ross Wightman 1420c118df Missed comitting outstanding changes to default_cfg keys and test exclusions for swin v2
3 years ago
Ross Wightman c6e4b7895a Swin V2 CR impl refactor.
3 years ago
Christoph Reich 67d140446b Fix bug in classification head
3 years ago
Christoph Reich 29add820ac Refactor (back to relative imports)
3 years ago
Christoph Reich 74a04e0016 Add parameter to change normalization type
3 years ago
Christoph Reich 2a4f6c13dd Create model functions
3 years ago
Christoph Reich 87b4d7a29a Add get and reset classifier method
3 years ago
Christoph Reich ff5f6bcd6c Check input resolution
3 years ago
Christoph Reich 81bf0b4033 Change parameter names to match Swin V1
3 years ago
Christoph Reich f227b88831 Add initials (CR) to model and file
3 years ago
Christoph Reich 90dc74c450 Add code from https://github.com/ChristophReich1996/Swin-Transformer-V2 and change docstring style to match timm
3 years ago
Ross Wightman 2c3870e107 semobilevit_s for good measure
3 years ago
Ross Wightman bcaeb91b03 Version to 0.6.0, possible interface incompatibilities vs 0.5.x
3 years ago
Ross Wightman 58ba49c8ef Add MobileViT models (w/ ByobNet base). Close #1038.
3 years ago
Ross Wightman 5f81d4de23 Move DeiT to own file, vit getting crowded. Working towards fixing #1029, make pooling interface for transformers and mlp closer to convnets. Still working through some details...
3 years ago
ayasyrev cf57695938 sched noise dup code remove
3 years ago
Ross Wightman 95cfc9b3e8 Merge remote-tracking branch 'origin/master' into norm_norm_norm
3 years ago
Ross Wightman abc9ba2544 Transitioning default_cfg -> pretrained_cfg. Improving handling of pretrained_cfg source (HF-Hub, files, timm config, etc). Checkpoint handling tweaks.
3 years ago
Ross Wightman 07379c6d5d Add vit_base2_patch32_256 for a model between base_patch16 and patch32 with a slightly larger img size and width
3 years ago
Ross Wightman 447677616f version 0.5.5
3 years ago
Ross Wightman 83b40c5a58 Last batch of small model weights (for now). mobilenetv3_small 050/075/100 and updated mnasnet_small with lambc/lamb optimizer.
3 years ago
Mi-Peng cdcd0a92ca fix lars
3 years ago
Ross Wightman 1aa617cb3b Add AvgPool2d anti-aliasing support to ResNet arch (as per OpenAI CLIP models), add a few blur aa models as well
3 years ago
Ross Wightman f0f9eccda8 Add --fuser arg to train/validate/benchmark scripts to select jit fuser type
3 years ago
Ross Wightman 010b486590 Add Dino pretrained weights (no head) for vit models. Add support to tests and helpers for models w/ no classifier (num_classes=0 in pretrained cfg)
3 years ago
Ross Wightman 738a9cd635 unbiased=False for torch.var_mean path of ConvNeXt LN. Fix #1090
3 years ago
Ross Wightman e0c4eec4b6 Default conv_mlp to False across the board for ConvNeXt, causing issues on more setups than it's improving right now...
3 years ago
Ross Wightman b669f4a588 Add ConvNeXt 22k->1k fine-tuned and 384 22k-1k fine-tuned weights after testing
3 years ago
Ross Wightman e967c72875 Update REAMDE.md. Sneak in g/G (giant / gigantic?) ViT defs from scaling paper
3 years ago
Ross Wightman 9ca3437178 Add some more small model weights lcnet, mnas, mnv2
3 years ago
Ross Wightman fa6463c936 Version 0.5.4
3 years ago
Ross Wightman fa81164378 Fix stem width for really small mobilenetv3 arch defs
3 years ago
Ross Wightman edd3d73695 Add missing dropout for head reset in ConvNeXt default head
3 years ago
Ross Wightman b093dcb46d Some convnext cleanup, remove in place mul_ for gamma, breaking symbolic trace, cleanup head a bit...
3 years ago
Ross Wightman 18934debc5 Add initial ConvNeXt impl (mods of official code)
3 years ago
Ross Wightman 656757d26b Fix MobileNetV2 head conv size for multiplier < 1.0. Add some missing modification copyrights, fix starting date of some old ones.
3 years ago
Ross Wightman ccfeb06936 Fix out_indices handling breakage, should have left as per vgg approach.
3 years ago
Ross Wightman a9f91483a6 Fix #1078, DarkNet has 6 feature maps. Make vgg and darknet out_indices handling/comments equivalent
3 years ago
Ross Wightman c21b21660d visformer supports spatial feat map, update pool_size in pretrained cfg to match
3 years ago
Ross Wightman 9c11dfd9cb Fix fbnetv3 pretrained cfg changes
3 years ago
Ross Wightman 1406cddc2e FBNetV3 timm trained weights added for b/d/g variants. Update version to 0.5.2 for pypi release.
3 years ago
Ross Wightman 02ae11e526 Leaving repeat aug sampler indices as tensor thrashes worker shared process memory
3 years ago
Ross Wightman 4df51f3932 Add lcnet_100 and mnasnet_small weights
3 years ago
Ross Wightman 5ccf682a8f Remove deprecated bn-tf train arg and create_model handler. Add evos/evob models back into fx test filter until norm_norm_norm branch merged.
3 years ago
Ross Wightman b9a715c86a Add more small model defs for MobileNetV3/V2/LCNet
3 years ago
Ross Wightman b27c21b09a Update drop_path and drop_block (fast impl) to be symbolically traceable, slightly faster
3 years ago
Ross Wightman 214c84a235 Disable use of timm nn.Linear wrapper since AMP autocast + torchscript use appears fixed
3 years ago
Ross Wightman 72b57163d1 Merge branch 'master' of https://github.com/mrT23/pytorch-image-models into mrT23-master
3 years ago
Ross Wightman de5fa791c6 Merge branch 'master' into norm_norm_norm
3 years ago
Ross Wightman 26ff57f953 Add more small model defs for MobileNetV3/V2/LCNet
3 years ago
Hyeongchan Kim a0b2657497
Use `torch.repeat_interleave()` to generate repeated indices faster (#1058)
3 years ago
Ross Wightman 450ac6a0f5 Post merge tinynet fixes for pool_size, feature extraction
3 years ago