mirrors
/
pytorch-image-models
mirror of https://github.com/huggingface/pytorch-image-models.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Update README, tweak fine-tune effv2 model names.

pull/636/head
Ross Wightman 2021-05-14 16:37:43 -07:00
parent c4f482a08b
commit 328249f11a
3 changed files with 39 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
README.md
View File

@ -23,6 +23,15 @@ I'm fortunate to be able to dedicate significant time and money of my own suppor
## What's New ## What's New
### May 14, 2021
* Add EfficientNet-V2 official model defs w/ ported weights from official [Tensorflow/Keras](https://github.com/google/automl/tree/master/efficientnetv2) impl.
* 1k trained variants: `tf_efficientnetv2_s/m/l`
* 21k trained variants: `tf_efficientnetv2_s/m/l_21k`
* 21k pretrained -> 1k fine-tuned: `tf_efficientnetv2_s/m/l_21ft1k`
* v2 models w/ v1 scaling: `tf_efficientnet_v2_b0` through `b3`
* Rename my prev V2 guess `efficientnet_v2s` -> `efficientnetv2_rw_s`
* Some blank `efficientnetv2_*` models in-place for future native PyTorch training
### May 5, 2021 ### May 5, 2021
* Add MLP-Mixer models and port pretrained weights from [Google JAX impl](https://github.com/google-research/vision_transformer/tree/linen) * Add MLP-Mixer models and port pretrained weights from [Google JAX impl](https://github.com/google-research/vision_transformer/tree/linen)
* Add CaiT models and pretrained weights from [FB](https://github.com/facebookresearch/deit) * Add CaiT models and pretrained weights from [FB](https://github.com/facebookresearch/deit)

48
timm/models/efficientnet.py
View File

@ -1,4 +1,4 @@
""" PyTorch EfficientNet Family """ The EfficientNet Family in PyTorch
An implementation of EfficienNet that covers variety of related models with efficient architectures: An implementation of EfficienNet that covers variety of related models with efficient architectures:
@ -25,6 +25,10 @@ An implementation of EfficienNet that covers variety of related models with effi
* And likely more... * And likely more...
The majority of the above models (EfficientNet*, MixNet, MnasNet) and original weights were made available
by Mingxing Tan, Quoc Le, and other members of their Google Brain team. Thanks for consistently releasing
the models and weights open source!
Hacked together by / Copyright 2021 Ross Wightman Hacked together by / Copyright 2021 Ross Wightman
""" """
from functools import partial from functools import partial
@ -328,16 +332,16 @@ default_cfgs = {
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5), mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0), input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0),
'tf_efficientnetv2_s_21kft1k': _cfg( 'tf_efficientnetv2_s_21ft1k': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_s_21kft1k-d7dafa41.pth', url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_s_21ft1k-d7dafa41.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5), mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 300, 300), test_input_size=(3, 384, 384), pool_size=(10, 10), crop_pct=1.0), input_size=(3, 300, 300), test_input_size=(3, 384, 384), pool_size=(10, 10), crop_pct=1.0),
'tf_efficientnetv2_m_21kft1k': _cfg( 'tf_efficientnetv2_m_21ft1k': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_m_21kft1k-bf41664a.pth', url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_m_21ft1k-bf41664a.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5), mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0), input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0),
'tf_efficientnetv2_l_21kft1k': _cfg( 'tf_efficientnetv2_l_21ft1k': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_l_21kft1k-60127a9d.pth', url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-effv2-weights/tf_efficientnetv2_l_21ft1k-60127a9d.pth',
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5), mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0), input_size=(3, 384, 384), test_input_size=(3, 480, 480), pool_size=(12, 12), crop_pct=1.0),
@ -1925,35 +1929,39 @@ def tf_efficientnetv2_l(pretrained=False, **kwargs):
@register_model @register_model
def tf_efficientnetv2_s_21kft1k(pretrained=False, **kwargs): def tf_efficientnetv2_s_21ft1k(pretrained=False, **kwargs):
""" EfficientNet-V2 Small. Tensorflow compatible variant """ """ EfficientNet-V2 Small. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_s('tf_efficientnetv2_s_21kft1k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_s('tf_efficientnetv2_s_21ft1k', pretrained=pretrained, **kwargs)
return model return model
@register_model @register_model
def tf_efficientnetv2_m_21kft1k(pretrained=False, **kwargs): def tf_efficientnetv2_m_21ft1k(pretrained=False, **kwargs):
""" EfficientNet-V2 Medium. Tensorflow compatible variant """ """ EfficientNet-V2 Medium. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_m('tf_efficientnetv2_m_21kft1k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_m('tf_efficientnetv2_m_21ft1k', pretrained=pretrained, **kwargs)
return model return model
@register_model @register_model
def tf_efficientnetv2_l_21kft1k(pretrained=False, **kwargs): def tf_efficientnetv2_l_21ft1k(pretrained=False, **kwargs):
""" EfficientNet-V2 Large. Tensorflow compatible variant """ """ EfficientNet-V2 Large. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_l('tf_efficientnetv2_l_21kft1k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_l('tf_efficientnetv2_l_21ft1k', pretrained=pretrained, **kwargs)
return model return model
@register_model @register_model
def tf_efficientnetv2_s_21k(pretrained=False, **kwargs): def tf_efficientnetv2_s_21k(pretrained=False, **kwargs):
""" EfficientNet-V2 Small w/ ImageNet-21k pretrained weights. Tensorflow compatible variant """ """ EfficientNet-V2 Small w/ ImageNet-21k pretrained weights. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_s('tf_efficientnetv2_s_21k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_s('tf_efficientnetv2_s_21k', pretrained=pretrained, **kwargs)
@ -1962,7 +1970,8 @@ def tf_efficientnetv2_s_21k(pretrained=False, **kwargs):
@register_model @register_model
def tf_efficientnetv2_m_21k(pretrained=False, **kwargs): def tf_efficientnetv2_m_21k(pretrained=False, **kwargs):
""" EfficientNet-V2 Medium w/ ImageNet-21k pretrained weights. Tensorflow compatible variant """ """ EfficientNet-V2 Medium w/ ImageNet-21k pretrained weights. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_m('tf_efficientnetv2_m_21k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_m('tf_efficientnetv2_m_21k', pretrained=pretrained, **kwargs)
@ -1971,7 +1980,8 @@ def tf_efficientnetv2_m_21k(pretrained=False, **kwargs):
@register_model @register_model
def tf_efficientnetv2_l_21k(pretrained=False, **kwargs): def tf_efficientnetv2_l_21k(pretrained=False, **kwargs):
""" EfficientNet-V2 Large w/ ImageNet-21k pretrained weights. Tensorflow compatible variant """ """ EfficientNet-V2 Large w/ ImageNet-21k pretrained weights. Tensorflow compatible variant
"""
kwargs['bn_eps'] = BN_EPS_TF_DEFAULT kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
kwargs['pad_type'] = 'same' kwargs['pad_type'] = 'same'
model = _gen_efficientnetv2_l('tf_efficientnetv2_l_21k', pretrained=pretrained, **kwargs) model = _gen_efficientnetv2_l('tf_efficientnetv2_l_21k', pretrained=pretrained, **kwargs)

2
timm/version.py
View File

@ -1 +1 @@
__version__ = '0.4.8' __version__ = '0.4.9'