Backbones

class torchok.models.backbones.base_backbone.BaseBackbone(in_channels: Optional[Union[int, List[int], Tuple[int, ...]]] = None, out_channels: Optional[Union[int, List[int], Tuple[int, ...]]] = None)

Bases: BaseModel, ABC

Base model for TorchOk Backbones

create_hooks(): Crete hooks for intermediate encoder features based on model’s feature info.

forward_features(x: Tensor) → List[Tensor]: Forward method for getting backbone feature maps. They are mainly used for segmentation and detection tasks.

property out_encoder_channels: Tuple[int]: Number of output feature channels - channels after forward_features method.

abstract get_stages(stage: int) → Module

Return modules corresponding the given model stage and all previous stages. For example, 0 must stand for model stem. 1 must stand for models stem and the first global layer of the model (layer1 in the resnet), etc.

Parameters: stage – index of the models stage.

training: bool

class torchok.models.backbones.base_backbone.BackboneWrapper(backbone)

Bases: Module

__init__(backbone)

forward(x)

property out_encoder_channels

training: bool

BEIT: BERT Pre-Training of Image Transformers (https://arxiv.org/abs/2106.08254)

Model from official source: https://github.com/microsoft/unilm/tree/master/beit

At this point only the 1k fine-tuned classification weights and model configs have been added, see original source above for pre-training models and procedure.

torchok.models.backbones.beit.beit_base_patch16_224(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_base_patch16_384(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_base_patch16_224_in22k(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_large_patch16_224(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_large_patch16_384(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_large_patch16_512(pretrained=False, **kwargs)

torchok.models.backbones.beit.beit_large_patch16_224_in22k(pretrained=False, **kwargs)

TorchOK DaViT.

Adapted from https://github.com/dingmyu/davit/blob/main/mmseg/mmseg/models/backbones/davit.py Licensed under MIT License [see LICENSE for details]

torchok.models.backbones.davit.davit_t(pretrained: bool = False, **kwargs): It’s constructing a davit_t model.

torchok.models.backbones.davit.davit_s(pretrained: bool = False, **kwargs): It’s constructing a davit_s model.

torchok.models.backbones.davit.davit_b(pretrained: bool = False, **kwargs): It’s constructing a davit_b model.

The EfficientNet Family in PyTorch Adapted from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/efficientnet.py

An implementation of EfficienNet that covers variety of related models with efficient architectures:

EfficientNet-V2 - EfficientNetV2: Smaller Models and Faster Training - https://arxiv.org/abs/2104.00298
EfficientNet (B0-B8, L2 + Tensorflow pretrained AutoAug/RandAug/AdvProp/NoisyStudent weight ports) - EfficientNet: Rethinking Model Scaling for CNNs - https://arxiv.org/abs/1905.11946 - CondConv: Conditionally Parameterized Convolutions for Efficient Inference - https://arxiv.org/abs/1904.04971 - Adversarial Examples Improve Image Recognition - https://arxiv.org/abs/1911.09665 - Self-training with Noisy Student improves ImageNet classification - https://arxiv.org/abs/1911.04252
MixNet (Small, Medium, and Large) - MixConv: Mixed Depthwise Convolutional Kernels - https://arxiv.org/abs/1907.09595
MNasNet B1, A1 (SE), Small - MnasNet: Platform-Aware Neural Architecture Search for Mobile - https://arxiv.org/abs/1807.11626
FBNet-C - FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable NAS - https://arxiv.org/abs/1812.03443
Single-Path NAS Pixel1 - Single-Path NAS: Designing Hardware-Efficient ConvNets - https://arxiv.org/abs/1904.02877
TinyNet
- Model Rubik’s Cube: Twisting Resolution, Depth and Width for TinyNets - https://arxiv.org/abs/2010.14819
- Definitions & weights borrowed from https://github.com/huawei-noah/CV-Backbones/tree/master/tinynet_pytorch
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!

torchok.models.backbones.efficientnet.mnasnet_050(pretrained=False, **kwargs): MNASNet B1, depth multiplier of 0.5.

torchok.models.backbones.efficientnet.mnasnet_075(pretrained=False, **kwargs): MNASNet B1, depth multiplier of 0.75.

torchok.models.backbones.efficientnet.mnasnet_100(pretrained=False, **kwargs): MNASNet B1, depth multiplier of 1.0.

torchok.models.backbones.efficientnet.mnasnet_b1(pretrained=False, **kwargs): MNASNet B1, depth multiplier of 1.0.

torchok.models.backbones.efficientnet.mnasnet_140(pretrained=False, **kwargs): MNASNet B1, depth multiplier of 1.4

torchok.models.backbones.efficientnet.semnasnet_050(pretrained=False, **kwargs): MNASNet A1 (w/ SE), depth multiplier of 0.5

torchok.models.backbones.efficientnet.semnasnet_075(pretrained=False, **kwargs): MNASNet A1 (w/ SE), depth multiplier of 0.75.

torchok.models.backbones.efficientnet.semnasnet_100(pretrained=False, **kwargs): MNASNet A1 (w/ SE), depth multiplier of 1.0.

torchok.models.backbones.efficientnet.mnasnet_a1(pretrained=False, **kwargs): MNASNet A1 (w/ SE), depth multiplier of 1.0.

torchok.models.backbones.efficientnet.semnasnet_140(pretrained=False, **kwargs): MNASNet A1 (w/ SE), depth multiplier of 1.4.

torchok.models.backbones.efficientnet.mnasnet_small(pretrained=False, **kwargs): MNASNet Small, depth multiplier of 1.0.

torchok.models.backbones.efficientnet.mobilenetv2_035(pretrained=False, **kwargs): MobileNet V2 w/ 0.35 channel multiplier

torchok.models.backbones.efficientnet.mobilenetv2_050(pretrained=False, **kwargs): MobileNet V2 w/ 0.5 channel multiplier

torchok.models.backbones.efficientnet.mobilenetv2_075(pretrained=False, **kwargs): MobileNet V2 w/ 0.75 channel multiplier

torchok.models.backbones.efficientnet.mobilenetv2_100(pretrained=False, **kwargs): MobileNet V2 w/ 1.0 channel multiplier

torchok.models.backbones.efficientnet.mobilenetv2_140(pretrained=False, **kwargs): MobileNet V2 w/ 1.4 channel multiplier

torchok.models.backbones.efficientnet.mobilenetv2_110d(pretrained=False, **kwargs): MobileNet V2 w/ 1.1 channel, 1.2 depth multipliers

torchok.models.backbones.efficientnet.mobilenetv2_120d(pretrained=False, **kwargs): MobileNet V2 w/ 1.2 channel, 1.4 depth multipliers

torchok.models.backbones.efficientnet.fbnetc_100(pretrained=False, **kwargs): FBNet-C

torchok.models.backbones.efficientnet.spnasnet_100(pretrained=False, **kwargs): Single-Path NAS Pixel1

torchok.models.backbones.efficientnet.efficientnet_b0(pretrained=False, **kwargs): EfficientNet-B0

torchok.models.backbones.efficientnet.efficientnet_b1(pretrained=False, **kwargs): EfficientNet-B1

torchok.models.backbones.efficientnet.efficientnet_b2(pretrained=False, **kwargs): EfficientNet-B2

torchok.models.backbones.efficientnet.efficientnet_b2a(pretrained=False, **kwargs): EfficientNet-B2 @ 288x288 w/ 1.0 test crop

torchok.models.backbones.efficientnet.efficientnet_b3(pretrained=False, **kwargs): EfficientNet-B3

torchok.models.backbones.efficientnet.efficientnet_b3a(pretrained=False, **kwargs): EfficientNet-B3 @ 320x320 w/ 1.0 test crop-pct

torchok.models.backbones.efficientnet.efficientnet_b4(pretrained=False, **kwargs): EfficientNet-B4

torchok.models.backbones.efficientnet.efficientnet_b5(pretrained=False, **kwargs): EfficientNet-B5

torchok.models.backbones.efficientnet.efficientnet_b6(pretrained=False, **kwargs): EfficientNet-B6

torchok.models.backbones.efficientnet.efficientnet_b7(pretrained=False, **kwargs): EfficientNet-B7

torchok.models.backbones.efficientnet.efficientnet_b8(pretrained=False, **kwargs): EfficientNet-B8

torchok.models.backbones.efficientnet.efficientnet_l2(pretrained=False, **kwargs): EfficientNet-L2.

torchok.models.backbones.efficientnet.efficientnet_b0_gn(pretrained=False, **kwargs): EfficientNet-B0 + GroupNorm

torchok.models.backbones.efficientnet.efficientnet_b0_g8_gn(pretrained=False, **kwargs): EfficientNet-B0 w/ group conv + GroupNorm

torchok.models.backbones.efficientnet.efficientnet_b0_g16_evos(pretrained=False, **kwargs): EfficientNet-B0 w/ group 16 conv + EvoNorm

torchok.models.backbones.efficientnet.efficientnet_b3_gn(pretrained=False, **kwargs): EfficientNet-B3 w/ GroupNorm

torchok.models.backbones.efficientnet.efficientnet_b3_g8_gn(pretrained=False, **kwargs): EfficientNet-B3 w/ grouped conv + BN

torchok.models.backbones.efficientnet.efficientnet_es(pretrained=False, **kwargs): EfficientNet-Edge Small.

torchok.models.backbones.efficientnet.efficientnet_em(pretrained=False, **kwargs): EfficientNet-Edge-Medium.

torchok.models.backbones.efficientnet.efficientnet_el(pretrained=False, **kwargs): EfficientNet-Edge-Large.

torchok.models.backbones.efficientnet.efficientnet_cc_b0_4e(pretrained=False, **kwargs): EfficientNet-CondConv-B0 w/ 8 Experts

torchok.models.backbones.efficientnet.efficientnet_cc_b0_8e(pretrained=False, **kwargs): EfficientNet-CondConv-B0 w/ 8 Experts

torchok.models.backbones.efficientnet.efficientnet_cc_b1_8e(pretrained=False, **kwargs): EfficientNet-CondConv-B1 w/ 8 Experts

torchok.models.backbones.efficientnet.efficientnet_lite0(pretrained=False, **kwargs): EfficientNet-Lite0

torchok.models.backbones.efficientnet.efficientnet_lite1(pretrained=False, **kwargs): EfficientNet-Lite1

torchok.models.backbones.efficientnet.efficientnet_lite2(pretrained=False, **kwargs): EfficientNet-Lite2

torchok.models.backbones.efficientnet.efficientnet_lite3(pretrained=False, **kwargs): EfficientNet-Lite3

torchok.models.backbones.efficientnet.efficientnet_lite4(pretrained=False, **kwargs): EfficientNet-Lite4

torchok.models.backbones.efficientnet.efficientnetv2_rw_t(pretrained=False, **kwargs): EfficientNet-V2 Tiny (Custom variant, tiny not in paper).

torchok.models.backbones.efficientnet.gc_efficientnetv2_rw_t(pretrained=False, **kwargs): EfficientNet-V2 Tiny w/ Global Context Attn (Custom variant, tiny not in paper).

torchok.models.backbones.efficientnet.efficientnetv2_rw_s(pretrained=False, **kwargs): EfficientNet-V2 Small (RW variant). NOTE: This is my initial (pre official code release) w/ some differences. See efficientnetv2_s and tf_efficientnetv2_s for versions that match the official w/ PyTorch vs TF padding

torchok.models.backbones.efficientnet.efficientnetv2_rw_m(pretrained=False, **kwargs): EfficientNet-V2 Medium (RW variant).

torchok.models.backbones.efficientnet.efficientnetv2_s(pretrained=False, **kwargs): EfficientNet-V2 Small.

torchok.models.backbones.efficientnet.efficientnetv2_m(pretrained=False, **kwargs): EfficientNet-V2 Medium.

torchok.models.backbones.efficientnet.efficientnetv2_l(pretrained=False, **kwargs): EfficientNet-V2 Large.

torchok.models.backbones.efficientnet.efficientnetv2_xl(pretrained=False, **kwargs): EfficientNet-V2 Xtra-Large.

torchok.models.backbones.efficientnet.tf_efficientnet_b0(pretrained=False, **kwargs): EfficientNet-B0. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b1(pretrained=False, **kwargs): EfficientNet-B1. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b2(pretrained=False, **kwargs): EfficientNet-B2. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b3(pretrained=False, **kwargs): EfficientNet-B3. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b4(pretrained=False, **kwargs): EfficientNet-B4. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b5(pretrained=False, **kwargs): EfficientNet-B5. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b6(pretrained=False, **kwargs): EfficientNet-B6. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b7(pretrained=False, **kwargs): EfficientNet-B7. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b8(pretrained=False, **kwargs): EfficientNet-B8. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b0_ap(pretrained=False, **kwargs): EfficientNet-B0 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b1_ap(pretrained=False, **kwargs): EfficientNet-B1 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b2_ap(pretrained=False, **kwargs): EfficientNet-B2 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b3_ap(pretrained=False, **kwargs): EfficientNet-B3 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b4_ap(pretrained=False, **kwargs): EfficientNet-B4 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b5_ap(pretrained=False, **kwargs): EfficientNet-B5 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b6_ap(pretrained=False, **kwargs): EfficientNet-B6 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b7_ap(pretrained=False, **kwargs): EfficientNet-B7 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b8_ap(pretrained=False, **kwargs): EfficientNet-B8 AdvProp. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b0_ns(pretrained=False, **kwargs): EfficientNet-B0 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b1_ns(pretrained=False, **kwargs): EfficientNet-B1 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b2_ns(pretrained=False, **kwargs): EfficientNet-B2 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b3_ns(pretrained=False, **kwargs): EfficientNet-B3 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b4_ns(pretrained=False, **kwargs): EfficientNet-B4 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b5_ns(pretrained=False, **kwargs): EfficientNet-B5 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b6_ns(pretrained=False, **kwargs): EfficientNet-B6 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_b7_ns(pretrained=False, **kwargs): EfficientNet-B7 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_l2_ns_475(pretrained=False, **kwargs): EfficientNet-L2 NoisyStudent @ 475x475. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_l2_ns(pretrained=False, **kwargs): EfficientNet-L2 NoisyStudent. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_es(pretrained=False, **kwargs): EfficientNet-Edge Small. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_em(pretrained=False, **kwargs): EfficientNet-Edge-Medium. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_el(pretrained=False, **kwargs): EfficientNet-Edge-Large. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_cc_b0_4e(pretrained=False, **kwargs): EfficientNet-CondConv-B0 w/ 4 Experts. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_cc_b0_8e(pretrained=False, **kwargs): EfficientNet-CondConv-B0 w/ 8 Experts. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_cc_b1_8e(pretrained=False, **kwargs): EfficientNet-CondConv-B1 w/ 8 Experts. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnet_lite0(pretrained=False, **kwargs): EfficientNet-Lite0

torchok.models.backbones.efficientnet.tf_efficientnet_lite1(pretrained=False, **kwargs): EfficientNet-Lite1

torchok.models.backbones.efficientnet.tf_efficientnet_lite2(pretrained=False, **kwargs): EfficientNet-Lite2

torchok.models.backbones.efficientnet.tf_efficientnet_lite3(pretrained=False, **kwargs): EfficientNet-Lite3

torchok.models.backbones.efficientnet.tf_efficientnet_lite4(pretrained=False, **kwargs): EfficientNet-Lite4

torchok.models.backbones.efficientnet.tf_efficientnetv2_s(pretrained=False, **kwargs): EfficientNet-V2 Small. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_m(pretrained=False, **kwargs): EfficientNet-V2 Medium. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_l(pretrained=False, **kwargs): EfficientNet-V2 Large. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_s_in21ft1k(pretrained=False, **kwargs): EfficientNet-V2 Small. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_m_in21ft1k(pretrained=False, **kwargs): EfficientNet-V2 Medium. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_l_in21ft1k(pretrained=False, **kwargs): EfficientNet-V2 Large. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_xl_in21ft1k(pretrained=False, **kwargs): EfficientNet-V2 Xtra-Large. Pretrained on ImageNet-21k, fine-tuned on 1k. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_s_in21k(pretrained=False, **kwargs): EfficientNet-V2 Small w/ ImageNet-21k pretrained weights. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_m_in21k(pretrained=False, **kwargs): EfficientNet-V2 Medium w/ ImageNet-21k pretrained weights. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_l_in21k(pretrained=False, **kwargs): EfficientNet-V2 Large w/ ImageNet-21k pretrained weights. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_xl_in21k(pretrained=False, **kwargs): EfficientNet-V2 Xtra-Large w/ ImageNet-21k pretrained weights. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_b0(pretrained=False, **kwargs): EfficientNet-V2-B0. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_b1(pretrained=False, **kwargs): EfficientNet-V2-B1. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_b2(pretrained=False, **kwargs): EfficientNet-V2-B2. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_efficientnetv2_b3(pretrained=False, **kwargs): EfficientNet-V2-B3. Tensorflow compatible variant

torchok.models.backbones.efficientnet.mixnet_s(pretrained=False, **kwargs): Creates a MixNet Small model.

torchok.models.backbones.efficientnet.mixnet_m(pretrained=False, **kwargs): Creates a MixNet Medium model.

torchok.models.backbones.efficientnet.mixnet_l(pretrained=False, **kwargs): Creates a MixNet Large model.

torchok.models.backbones.efficientnet.mixnet_xl(pretrained=False, **kwargs): Creates a MixNet Extra-Large model. Not a paper spec, experimental def by RW w/ depth scaling.

torchok.models.backbones.efficientnet.mixnet_xxl(pretrained=False, **kwargs): Creates a MixNet Double Extra Large model. Not a paper spec, experimental def by RW w/ depth scaling.

torchok.models.backbones.efficientnet.tf_mixnet_s(pretrained=False, **kwargs): Creates a MixNet Small model. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_mixnet_m(pretrained=False, **kwargs): Creates a MixNet Medium model. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tf_mixnet_l(pretrained=False, **kwargs): Creates a MixNet Large model. Tensorflow compatible variant

torchok.models.backbones.efficientnet.tinynet_a(pretrained=False, **kwargs)

torchok.models.backbones.efficientnet.tinynet_b(pretrained=False, **kwargs)

torchok.models.backbones.efficientnet.tinynet_c(pretrained=False, **kwargs)

torchok.models.backbones.efficientnet.tinynet_d(pretrained=False, **kwargs)

torchok.models.backbones.efficientnet.tinynet_e(pretrained=False, **kwargs)

TorchOK HRNet.

torchok.models.backbones.hrnet.hrnet_w18_small(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w18_small model.

torchok.models.backbones.hrnet.hrnet_w18_small_v2(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w18_small_v2 model.

torchok.models.backbones.hrnet.hrnet_w18(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w18 model.

torchok.models.backbones.hrnet.hrnet_w30(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w30 model.

torchok.models.backbones.hrnet.hrnet_w32(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w32 model.

torchok.models.backbones.hrnet.hrnet_w40(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w40 model.

torchok.models.backbones.hrnet.hrnet_w44(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w44 model.

torchok.models.backbones.hrnet.hrnet_w48(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w48 model.

torchok.models.backbones.hrnet.hrnet_w64(pretrained: bool = False, **kwargs): It’s constructing a hrnet_w64 model.

MobileNet V3 Adapted from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/mobilenetv3.py A PyTorch impl of MobileNet-V3, compatible with TF weights from official impl.

Paper: Searching for MobileNetV3 - https://arxiv.org/abs/1905.02244

torchok.models.backbones.mobilenetv3.mobilenetv3_large_075(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.mobilenetv3_large_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.mobilenetv3_large_100_miil(pretrained=False, **kwargs): MobileNet V3 Weights taken from: https://github.com/Alibaba-MIIL/ImageNet21K

torchok.models.backbones.mobilenetv3.mobilenetv3_large_100_miil_in21k(pretrained=False, **kwargs): MobileNet V3, 21k pretraining Weights taken from: https://github.com/Alibaba-MIIL/ImageNet21K

torchok.models.backbones.mobilenetv3.mobilenetv3_small_050(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.mobilenetv3_small_075(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.mobilenetv3_small_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.mobilenetv3_rw(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_large_075(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_large_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_large_minimal_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_small_075(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_small_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.tf_mobilenetv3_small_minimal_100(pretrained=False, **kwargs): MobileNet V3

torchok.models.backbones.mobilenetv3.fbnetv3_b(pretrained=False, **kwargs): FBNetV3-B

torchok.models.backbones.mobilenetv3.fbnetv3_d(pretrained=False, **kwargs): FBNetV3-D

torchok.models.backbones.mobilenetv3.fbnetv3_g(pretrained=False, **kwargs): FBNetV3-G

torchok.models.backbones.mobilenetv3.lcnet_035(pretrained=False, **kwargs): PP-LCNet 0.35

torchok.models.backbones.mobilenetv3.lcnet_050(pretrained=False, **kwargs): PP-LCNet 0.5

torchok.models.backbones.mobilenetv3.lcnet_075(pretrained=False, **kwargs): PP-LCNet 1.0

torchok.models.backbones.mobilenetv3.lcnet_100(pretrained=False, **kwargs): PP-LCNet 1.0

torchok.models.backbones.mobilenetv3.lcnet_150(pretrained=False, **kwargs): PP-LCNet 1.5

TorchOK ResNet.

torchok.models.backbones.resnet.resnet10t(pretrained=False, **kwargs): Constructs a ResNet-10-T model.

torchok.models.backbones.resnet.resnet14t(pretrained=False, **kwargs): Constructs a ResNet-14-T model.

torchok.models.backbones.resnet.resnet18(pretrained=False, **kwargs): Constructs a ResNet-18 model.

torchok.models.backbones.resnet.resnet18d(pretrained=False, **kwargs): Constructs a ResNet-18-D model.

torchok.models.backbones.resnet.resnet34(pretrained=False, **kwargs): Constructs a ResNet-34 model.

torchok.models.backbones.resnet.resnet34d(pretrained=False, **kwargs): Constructs a ResNet-34-D model.

torchok.models.backbones.resnet.resnet26(pretrained=False, **kwargs): Constructs a ResNet-26 model.

torchok.models.backbones.resnet.resnet26t(pretrained=False, **kwargs): Constructs a ResNet-26-T model.

torchok.models.backbones.resnet.resnet26d(pretrained=False, **kwargs): Constructs a ResNet-26-D model.

torchok.models.backbones.resnet.resnet50(pretrained=False, **kwargs): Constructs a ResNet-50 model.

torchok.models.backbones.resnet.resnet50d(pretrained=False, **kwargs): Constructs a ResNet-50-D model.

torchok.models.backbones.resnet.resnet50t(pretrained=False, **kwargs): Constructs a ResNet-50-T model.

torchok.models.backbones.resnet.resnet101(pretrained=False, **kwargs): Constructs a ResNet-101 model.

torchok.models.backbones.resnet.resnet101d(pretrained=False, **kwargs): Constructs a ResNet-101-D model.

torchok.models.backbones.resnet.resnet152(pretrained=False, **kwargs): Constructs a ResNet-152 model.

torchok.models.backbones.resnet.resnet152d(pretrained=False, **kwargs): Constructs a ResNet-152-D model.

torchok.models.backbones.resnet.resnet200(pretrained=False, **kwargs): Constructs a ResNet-200 model.

torchok.models.backbones.resnet.resnet200d(pretrained=False, **kwargs): Constructs a ResNet-200-D model.

torchok.models.backbones.resnet.tv_resnet34(pretrained=False, **kwargs): Constructs a ResNet-34 model with original Torchvision weights.

torchok.models.backbones.resnet.tv_resnet50(pretrained=False, **kwargs): Constructs a ResNet-50 model with original Torchvision weights.

torchok.models.backbones.resnet.tv_resnet101(pretrained=False, **kwargs): Constructs a ResNet-101 model w/ Torchvision pretrained weights.

torchok.models.backbones.resnet.tv_resnet152(pretrained=False, **kwargs): Constructs a ResNet-152 model w/ Torchvision pretrained weights.

torchok.models.backbones.resnet.wide_resnet50_2(pretrained=False, **kwargs): Constructs a Wide ResNet-50-2 model. The model is the same as ResNet except for the bottleneck number of channels which is twice larger in every block. The number of channels in outer 1x1 convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048 channels, and in Wide ResNet-50-2 has 2048-1024-2048.

torchok.models.backbones.resnet.wide_resnet101_2(pretrained=False, **kwargs): Constructs a Wide ResNet-101-2 model. The model is the same as ResNet except for the bottleneck number of channels which is twice larger in every block. The number of channels in outer 1x1 convolutions is the same.

torchok.models.backbones.resnet.resnet50_gn(pretrained=False, **kwargs): Constructs a ResNet-50 model w/ GroupNorm

torchok.models.backbones.resnet.resnext50_32x4d(pretrained=False, **kwargs): Constructs a ResNeXt50-32x4d model.

torchok.models.backbones.resnet.resnext50d_32x4d(pretrained=False, **kwargs): Constructs a ResNeXt50d-32x4d model. ResNext50 w/ deep stem & avg pool downsample

torchok.models.backbones.resnet.resnext101_32x4d(pretrained=False, **kwargs): Constructs a ResNeXt-101 32x4d model.

torchok.models.backbones.resnet.resnext101_32x8d(pretrained=False, **kwargs): Constructs a ResNeXt-101 32x8d model.

torchok.models.backbones.resnet.resnext101_64x4d(pretrained=False, **kwargs): Constructs a ResNeXt101-64x4d model.

torchok.models.backbones.resnet.tv_resnext50_32x4d(pretrained=False, **kwargs): Constructs a ResNeXt50-32x4d model with original Torchvision weights.

torchok.models.backbones.resnet.ig_resnext101_32x8d(pretrained=True, **kwargs): Constructs a ResNeXt-101 32x8 model pre-trained on weakly-supervised data and finetuned on ImageNet from Figure 5 in “Exploring the Limits of Weakly Supervised Pretraining” Weights from https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/

torchok.models.backbones.resnet.ig_resnext101_32x16d(pretrained=True, **kwargs): Constructs a ResNeXt-101 32x16 model pre-trained on weakly-supervised data and finetuned on ImageNet from Figure 5 in “Exploring the Limits of Weakly Supervised Pretraining” Weights from https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/

torchok.models.backbones.resnet.ig_resnext101_32x32d(pretrained=True, **kwargs): Constructs a ResNeXt-101 32x32 model pre-trained on weakly-supervised data and finetuned on ImageNet from Figure 5 in “Exploring the Limits of Weakly Supervised Pretraining” Weights from https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/

torchok.models.backbones.resnet.ig_resnext101_32x48d(pretrained=True, **kwargs): Constructs a ResNeXt-101 32x48 model pre-trained on weakly-supervised data and finetuned on ImageNet from Figure 5 in “Exploring the Limits of Weakly Supervised Pretraining” Weights from https://pytorch.org/hub/facebookresearch_WSL-Images_resnext/

torchok.models.backbones.resnet.ssl_resnet18(pretrained=True, **kwargs): Constructs a semi-supervised ResNet-18 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ssl_resnet50(pretrained=True, **kwargs): Constructs a semi-supervised ResNet-50 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ssl_resnext50_32x4d(pretrained=True, **kwargs): Constructs a semi-supervised ResNeXt-50 32x4 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ssl_resnext101_32x4d(pretrained=True, **kwargs): Constructs a semi-supervised ResNeXt-101 32x4 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ssl_resnext101_32x8d(pretrained=True, **kwargs): Constructs a semi-supervised ResNeXt-101 32x8 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ssl_resnext101_32x16d(pretrained=True, **kwargs): Constructs a semi-supervised ResNeXt-101 32x16 model pre-trained on YFCC100M dataset and finetuned on ImageNet “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnet18(pretrained=True, **kwargs): Constructs a semi-weakly supervised Resnet-18 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnet50(pretrained=True, **kwargs): Constructs a semi-weakly supervised ResNet-50 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnext50_32x4d(pretrained=True, **kwargs): Constructs a semi-weakly supervised ResNeXt-50 32x4 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnext101_32x4d(pretrained=True, **kwargs): Constructs a semi-weakly supervised ResNeXt-101 32x4 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnext101_32x8d(pretrained=True, **kwargs): Constructs a semi-weakly supervised ResNeXt-101 32x8 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.swsl_resnext101_32x16d(pretrained=True, **kwargs): Constructs a semi-weakly supervised ResNeXt-101 32x16 model pre-trained on 1B weakly supervised image dataset and finetuned on ImageNet. “Billion-scale Semi-Supervised Learning for Image Classification” Weights from https://github.com/facebookresearch/semi-supervised-ImageNet1K-models/

torchok.models.backbones.resnet.ecaresnet26t(pretrained=False, **kwargs): Constructs an ECA-ResNeXt-26-T model. This is technically a 28 layer ResNet, like a ‘D’ bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem and ECA attn.

torchok.models.backbones.resnet.ecaresnet50d(pretrained=False, **kwargs): Constructs a ResNet-50-D model with eca.

torchok.models.backbones.resnet.ecaresnet50t(pretrained=False, **kwargs): Constructs an ECA-ResNet-50-T model. Like a ‘D’ bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem and ECA attn.

torchok.models.backbones.resnet.ecaresnetlight(pretrained=False, **kwargs): Constructs a ResNet-50-D light model with eca.

torchok.models.backbones.resnet.ecaresnet101d(pretrained=False, **kwargs): Constructs a ResNet-101-D model with eca.

torchok.models.backbones.resnet.ecaresnet200d(pretrained=False, **kwargs): Constructs a ResNet-200-D model with ECA.

torchok.models.backbones.resnet.ecaresnet269d(pretrained=False, **kwargs): Constructs a ResNet-269-D model with ECA.

torchok.models.backbones.resnet.ecaresnext26t_32x4d(pretrained=False, **kwargs): Constructs an ECA-ResNeXt-26-T model. This is technically a 28 layer ResNet, like a ‘D’ bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem. This model replaces SE module with the ECA module

torchok.models.backbones.resnet.ecaresnext50t_32x4d(pretrained=False, **kwargs): Constructs an ECA-ResNeXt-50-T model. This is technically a 28 layer ResNet, like a ‘D’ bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem. This model replaces SE module with the ECA module

torchok.models.backbones.resnet.seresnet18(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet34(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet50(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet50t(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet101(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet152(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet152d(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnet200d(pretrained=False, **kwargs): Constructs a ResNet-200-D model with SE attn.

torchok.models.backbones.resnet.seresnet269d(pretrained=False, **kwargs): Constructs a ResNet-269-D model with SE attn.

torchok.models.backbones.resnet.seresnext26d_32x4d(pretrained=False, **kwargs): Constructs a SE-ResNeXt-26-D model.` This is technically a 28 layer ResNet, using the ‘D’ modifier from Gluon / bag-of-tricks for combination of deep stem and avg_pool in downsample.

torchok.models.backbones.resnet.seresnext26t_32x4d(pretrained=False, **kwargs): Constructs a SE-ResNet-26-T model. This is technically a 28 layer ResNet, like a ‘D’ bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem.

torchok.models.backbones.resnet.seresnext26tn_32x4d(pretrained=False, **kwargs): Constructs a SE-ResNeXt-26-T model. NOTE I deprecated previous ‘t’ model defs and replaced ‘t’ with ‘tn’, this was the only tn model of note so keeping this def for backwards compat with any uses out there. Old ‘t’ model is lost.

torchok.models.backbones.resnet.seresnext50_32x4d(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnext101_32x4d(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnext101_32x8d(pretrained=False, **kwargs)

torchok.models.backbones.resnet.seresnext101d_32x8d(pretrained=False, **kwargs)

torchok.models.backbones.resnet.senet154(pretrained=False, **kwargs)

torchok.models.backbones.resnet.resnetblur18(pretrained=False, **kwargs): Constructs a ResNet-18 model with blur anti-aliasing

torchok.models.backbones.resnet.resnetblur50(pretrained=False, **kwargs): Constructs a ResNet-50 model with blur anti-aliasing

torchok.models.backbones.resnet.resnetblur50d(pretrained=False, **kwargs): Constructs a ResNet-50-D model with blur anti-aliasing

torchok.models.backbones.resnet.resnetblur101d(pretrained=False, **kwargs): Constructs a ResNet-101-D model with blur anti-aliasing

torchok.models.backbones.resnet.resnetaa50(pretrained=False, **kwargs): Constructs a ResNet-50 model with avgpool anti-aliasing

torchok.models.backbones.resnet.resnetaa50d(pretrained=False, **kwargs): Constructs a ResNet-50-D model with avgpool anti-aliasing

torchok.models.backbones.resnet.resnetaa101d(pretrained=False, **kwargs): Constructs a ResNet-101-D model with avgpool anti-aliasing

torchok.models.backbones.resnet.seresnetaa50d(pretrained=False, **kwargs): Constructs a SE=ResNet-50-D model with avgpool anti-aliasing

torchok.models.backbones.resnet.seresnextaa101d_32x8d(pretrained=False, **kwargs): Constructs a SE=ResNeXt-101-D 32x8d model with avgpool anti-aliasing

torchok.models.backbones.resnet.resnetrs50(pretrained=False, **kwargs): Constructs a ResNet-RS-50 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs101(pretrained=False, **kwargs): Constructs a ResNet-RS-101 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs152(pretrained=False, **kwargs): Constructs a ResNet-RS-152 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs200(pretrained=False, **kwargs): Constructs a ResNet-RS-200 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs270(pretrained=False, **kwargs): Constructs a ResNet-RS-270 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs350(pretrained=False, **kwargs): Constructs a ResNet-RS-350 model. Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

torchok.models.backbones.resnet.resnetrs420(pretrained=False, **kwargs): Constructs a ResNet-RS-420 model Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579 Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs

TorchOK Swin Transformer V2 A PyTorch implementation of Swin Transformer V2: Scaling Up Capacity and Resolution - https://arxiv.org/abs/2111.09883

Adapted from https://github.com/microsoft/Swin-Transformer/blob/main/models/swin_transformer_v2.py (Copyright (c) 2022 Microsoft) and from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/swin_transformer_v2.py (Copyright 2022, Ross Wightman) Licensed under Apache License 2.0 [see LICENSE for details]

torchok.models.backbones.swin.swinv2_custom(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_tiny_window16_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_tiny_window8_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_small_window16_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_small_window8_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_base_window16_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_base_window8_256(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_base_window12_192_22k(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_base_window12to16_192to256_22kft1k(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_base_window12to24_192to384_22kft1k(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_large_window12_192_22k(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_large_window12to16_192to256_22kft1k(pretrained=False, **kwargs)

torchok.models.backbones.swin.swinv2_large_window12to24_192to384_22kft1k(pretrained=False, **kwargs)