05313940e2
Add cs3darknet_x, cs3sedarknet_l, and darknetaa53 weights from TPU sessions. Move SE btwn conv1 & conv2 in DarkBlock. Improve SE/attn handling in Csp/DarkNet. Fix leaky_relu bug on older csp models.
2022-07-15 16:55:16 -07:00
51cca82aa1
👽 use hf_hub_download instead of cached_download
2022-07-14 16:41:45 -04:00
a45b4bce9a
x and xx small edgenext models do benefit from larger test input size
2022-07-08 10:53:27 -07:00
a8e34051c1
Unbreak gamma remap impacting beit checkpoint load, version bump to 0.6.4
2022-07-07 23:07:43 -07:00
a1cb25066e
Add edgnext_small_rw weights trained with swin like recipe. Better than original 'small' but not the recent 'USI' distilled weights.
2022-07-07 22:02:57 -07:00
7c7ecd2492
Add --use-train-size flag to force use of train input_size (over test input size) for validation. Default test-time pooling to use train input size (fixes issues).
2022-07-07 22:01:24 -07:00
ce65a7b29f
Update vit_relpos w/ some additional weights, some cleanup to match recent vit updates, more MLP log coord experiments.
2022-07-07 21:33:25 -07:00
58621723bd
Add CrossStage3 DarkNet (cs3) weights
2022-07-07 17:43:38 -07:00
db0cee9910
Refactor cspnet configuration using dataclasses, update feature extraction for new cs3 variants.
2022-07-07 14:43:27 -07:00
eca09b8642
Add MobileVitV2 support. Fix #1332 . Move GroupNorm1 to common layers (used in poolformer + mobilevitv2). Keep ol custom ConvNeXt LayerNorm2d impl as LayerNormExp2d for reference.
2022-07-07 14:41:01 -07:00
06307b8b41
Remove experimental downsample in block support in ConvNeXt. Experiment further before keeping it in.
2022-07-07 14:37:58 -07:00
7d4b3807d5
Support DeiT-3 (Revenge of the ViT) checkpoints. Add non-overlapping (w/ class token) pos-embed support to vit.
2022-07-04 22:25:22 -07:00
d0c5bd5722
Rename cs2->cs3 for darknets. Fix features_only for cs3 darknets.
2022-07-03 08:32:41 -07:00
d765305821
Remove first_conv for resnetaa50 def
2022-07-02 15:56:17 -07:00
dd9b8f57c4
Add feature_info to edgenext for features_only support, hopefully fix some fx / test errors
2022-07-02 15:20:45 -07:00
377e9bfa21
Add TPU trained darknet53 weights. Add mising pretrain_cfg for some csp/darknet models.
2022-07-02 15:18:52 -07:00
c170ba3173
Add weights for resnet10t, resnet14t, and resnetaa50 models. Fix #1314
2022-07-02 15:18:06 -07:00
188c194b0f
Left some experiment stem code in convnext by mistake
2022-07-02 15:17:28 -07:00
6064d16a2d
Add initial EdgeNeXt import. Significant cleanup / reorg (like ConvNeXt). Fix #1320
...
* edgenext refactored for torchscript compat, stage base organization
* slight refactor of ConvNeXt to match some EdgeNeXt additions
* remove use of funky LayerNorm layer in ConvNeXt and just use nn.LayerNorm and LayerNorm2d (permute)
2022-07-01 15:18:42 -07:00
7a9c6811c9
Add eps arg to LayerNorm2d, add 'tf' (tensorflow) variant of trunc_normal_ that applies scale/shift after sampling (instead of needing to move a/b)
2022-07-01 15:15:39 -07:00
82c311d082
Add more experimental darknet and 'cs2' darknet variants (different cross stage setup, closer to newer YOLO backbones) for train trials.
2022-07-01 15:14:01 -07:00
a050fde5cd
Add resnet10t (basic block) and resnet14t (bottleneck) with 1,1,1,1 repeats
2022-07-01 15:03:28 -07:00
e6d7df40ec
no longer a point using kwargs for pretrain_cfg resolve, just pass explicit arg
2022-06-24 21:36:23 -07:00
07d0c4ae96
Improve repr for DropPath module
2022-06-24 14:58:15 -07:00
e27c16b8a0
Remove unecessary code for synbn guard
2022-06-24 14:57:42 -07:00
0da3c9ebbf
Remove SiLU layer in default args that breaks import on old old PyTorch
2022-06-24 14:56:58 -07:00
7d657d2ef4
Improve resolve_pretrained_cfg behaviour when no cfg exists, warn instead of crash. Improve usability ex #1311
2022-06-24 14:55:25 -07:00
879df47c0a
Support BatchNormAct2d for sync-bn use. Fix #1254
2022-06-24 14:51:26 -07:00
4b30bae67b
Add updated vit_relpos weights, and impl w/ support for official swin-v2 differences for relpos. Add bias control support for MLP layers
2022-05-13 13:53:57 -07:00
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
2022-05-13 10:50:59 -07:00
27c42f0830
Fix torchscript use for offician Swin-V2, add support for non-square window/shift to WindowAttn/Block
2022-05-13 09:29:33 -07:00
c0211b0bf7
Swin-V2 test fixes, typo
2022-05-12 22:31:55 -07:00
9a86b900fa
Official SwinV2 models
2022-05-12 15:05:10 -07:00
d07d015173
Merge pull request #1249 from okojoalg/sequencer
...
Add Sequencer
2022-05-09 20:42:43 -07:00
39b725e1c9
Fix tests for rank-4 output where feature channels dim is -1 (3) and not 1
2022-05-09 15:20:24 -07:00
78a32655fa
Fix poolformer group_matcher to merge proj downsample with previous block, support coarse
2022-05-09 12:20:04 -07:00
d79f3d9d1e
Fix torchscript use for sequencer, add group_matcher, forward_head support, minor formatting
2022-05-09 12:09:39 -07:00
37b6920df3
Fix group_matcher regex for regnet.py
2022-05-09 10:40:40 -07:00
93a79a3dd9
Fix num_features in Sequencer
2022-05-06 23:16:32 +09:00
578d52e752
Add Sequencer
2022-05-06 00:36:01 +09:00
f5ca4141f7
Adjust arg order for recent vit model args, add a few comments
2022-05-02 22:41:38 -07:00
41dc49a337
Vision Transformer refactoring and Rel Pos impl
2022-05-02 15:37:39 -07:00
b7cb8d0337
Add Swin-V2 Small-NS weights (83.5 @ 224). Add layer scale like 'init_values' via post-norm LN weight scaling
2022-04-26 17:32:49 -07:00
f88c606fcf
fixing channels_last on cond_conv2d; update nvfuser debug env variable
2022-04-25 12:41:46 -07:00
09e9f3defb
migrate azure blob for beit checkpoints
...
## Motivation
We are going to use a new blob account to store the checkpoints.
## Modification
Modify the azure blob storage URLs for BEiT checkpoints.
2022-04-23 13:02:29 +08:00
52ac881402
Missed first_conv in latest seresnext 'D' default_cfgs
2022-04-22 20:55:52 -07:00
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.
2022-04-22 16:54:53 -07:00
8f0bc0591e
fix convnext args
2022-04-05 20:00:57 +08:00
c5a8e929fb
Add initial swinv2 tiny / small weights
2022-04-03 15:22:55 -07:00
f670d98cb8
Make a few more layers symbolically traceable (remove from FX leaf modules)
...
* remove dtype kwarg from .to() calls in EvoNorm as it messed up script + trace combo
* BatchNormAct2d always uses custom forward (cut & paste from original) instead of super().forward. Fixes #1176
* BlurPool groups==channels, no need to use input.dim[1]
2022-03-24 21:43:56 -07:00
Ross Wightman
nateraw
Ross Wightman
okojoalg
jjsjann123
Li Dong
SeeFun