sdkagent

pytorch-image-models API reference

400 public APIs from pytorch-image-models (huggingface/pytorch-image-models) — 131 classes, 196 functions, 73 methods. Signatures extracted by static analysis of the actual source.

Repository: huggingface/pytorch-image-models

KindCount
Classes131
Functions196
Methods73

API list

funcconvert.convert_gemma4_vit.remap_key(key:str) -> str | None
Remap a single HF Transformers key to timm format.
classtimm.data.readers.reader_wds.ResampledShards2
An iterable dataset yielding a list of urls.
classtimm.data.transforms.CenterCropOrPad
Crops the given image at the center.
classtimm.data.transforms.ResizeKeepRatio
Resize and Keep Aspect Ratio
classtimm.data.transforms.ToTensor
ToTensor with no rescaling of values
classtimm.layers.activations.PReLU
Applies PReLU (w/ dummy inplace arg)
classtimm.layers.activations_me.HardSwishAutoFn
A memory efficient HardSwish activation
classtimm.layers.attention.AttentionRope
A Self Attention module with ROPE support.
classtimm.layers.attention2d.MultiQueryAttention2d
Multi Query Attention with spatial downsampling.
methodtimm.layers.attention2d.MultiQueryAttention2d.forward(x, attn_mask:Optional[torch.Tensor]=None)
Run layer computation.
classtimm.layers.attention2d.MultiQueryAttentionV2
Multi Query Attention.
methodtimm.layers.attention2d.MultiQueryAttentionV2.forward(x, m:Optional[torch.Tensor]=None)
Run layer computation.
classtimm.layers.cbam.SpatialAttn
Original CBAM spatial attention module
classtimm.layers.coord_attn.EfficientLocalAttn
Efficient Local Attention.
classtimm.layers.coord_attn.SimpleCoordAttn
Simplified Coordinate Attention variant.
classtimm.layers.coord_attn.StripAttn
Minimal Strip Attention.
classtimm.layers.diff_attention.DiffAttention
Differential Attention module.
classtimm.layers.drop.DropBlock2d
DropBlock.
functimm.layers.drop.drop_block_2d(x:torch.Tensor, drop_prob:float=0.1, block_size:int=7, gamma_scale:float=1.0, with_noise:bool=False, inplace:bool=False, couple_channels:bool=True, scale_by_keep:bool=True)
DropBlock.
classtimm.layers.eca.CecaModule
Constructs a circular ECA module.
classtimm.layers.eca.EcaModule
Constructs an ECA module.
classtimm.layers.gather_excite.GatherExcite
Gather-Excite Attention Module
functimm.layers.grid.meshgrid(*tensors) -> Tuple[torch.Tensor, ...]
generate N-D grid in spatial dim order.
functimm.layers.grid.ndgrid(*tensors) -> Tuple[torch.Tensor, ...]
generate N-D grid in dimension order.
classtimm.layers.grn.GlobalResponseNorm
Global Response Normalization layer
classtimm.layers.layer_scale.LayerScale
LayerScale on tensors with channels in last-dim.
classtimm.layers.layer_scale.LayerScale2d
LayerScale for tensors with torch 2D NCHW layout.
classtimm.layers.mlp.GatedMlp
MLP as used in gMLP
classtimm.layers.non_local_attn.NonLocalAttn
Spatial NL block for image classification.
classtimm.layers.norm.GroupNorm1
Group Normalization with 1 group.
classtimm.layers.norm.LayerNorm
LayerNorm w/ fast norm option
classtimm.layers.norm.LayerNorm2d
LayerNorm for channels of '2D' spatial NCHW tensors
classtimm.layers.norm.LayerNorm2dFp32
LayerNorm for channels of '2D' spatial NCHW tensors
classtimm.layers.norm.LayerNormFp32
LayerNorm
classtimm.layers.norm.RmsNorm
RmsNorm w/ fast (apex) norm if available
classtimm.layers.norm.RmsNormFp32
RmsNorm w/ fast (apex) norm if available
classtimm.layers.norm.SimpleNorm
SimpleNorm (x / std(x))
classtimm.layers.norm.SimpleNorm2d
SimpleNorm for NCHW tensors
classtimm.layers.norm.SimpleNorm2dFp32
SimpleNorm for NCHW tensors
classtimm.layers.norm.SimpleNormFp32
SimpleNorm (x / std(x))
classtimm.layers.other_pool.LsePlus1d
LogSumExp (LSE) Pooling for sequence (NLC) inputs.
classtimm.layers.other_pool.LsePlus2d
LogSumExp (LSE) Pooling for 2D inputs.
classtimm.layers.patch_dropout.PatchDropout
Patch Dropout without returning indices.
classtimm.layers.patch_embed.PatchEmbed
2D Image to Patch Embedding
classtimm.layers.patch_embed.PatchEmbedWithSize
2D Image to Patch Embedding
classtimm.layers.pool2d_same.AvgPool2dSame
Tensorflow like 'SAME' wrapper for 2D average pooling
classtimm.layers.pool2d_same.MaxPool2dSame
Tensorflow like 'SAME' wrapper for 2D max pooling
methodtimm.layers.pos_embed_sincos.RotaryEmbedding.init_non_persistent_buffers() -> None
Initialize non-persistent buffers.
methodtimm.layers.pos_embed_sincos.RotaryEmbedding.reset_parameters() -> None
Initialize parameters and buffers.
functimm.layers.pos_embed_sincos.get_mixed_freqs(freqs:torch.Tensor, t_x:torch.Tensor, t_y:torch.Tensor) -> torch.Tensor
Compute mixed (learnable) frequencies.
classtimm.layers.separable_conv.SeparableConv2d
Separable Conv
classtimm.layers.separable_conv.SeparableConvNormAct
Separable Conv w/ trailing Norm and Activation
classtimm.layers.space_to_depth.DepthToSpace
Rearrange channel dimension into spatial dimensions.
classtimm.layers.space_to_depth.SpaceToDepth
Rearrange spatial dimensions into channel dimension.
classtimm.layers.split_attn.SplitAttn
Split-Attention (aka Splat)
classtimm.layers.std_conv.ScaledStdConv2d
Conv2d layer with Scaled Weight Standardization.
classtimm.layers.std_conv.StdConv2d
Conv2d with Weight Standardization.
classtimm.layers.std_conv.StdConv2dSame
Conv2d with Weight Standardization.
classtimm.loss.cross_entropy.LabelSmoothingCrossEntropy
NLL loss with label smoothing.
functimm.models._builder.set_pretrained_check_hash(enable:bool=True) -> None
Set hash checking for pretrained weights on/off (globally).
classtimm.models._efficientnet_blocks.CondConvResidual
Inverted residual block w/ CondConv routing
functimm.models._factory.parse_model_name(model_name:str) -> Tuple[Optional[str], str]
Parse source and name from potentially prefixed model name.
functimm.models._factory.safe_model_name(model_name:str, remove_source:bool=True) -> str
Return a filename / path safe model name.
functimm.models._registry.get_arch_pretrained_cfgs(model_name:str) -> Dict[str, PretrainedCfg]
Get all pretrained cfgs for a given architecture.
functimm.models._registry.get_pretrained_cfg_value(model_name:str, cfg_key:str) -> Optional[Any]
Get a specific model default_cfg value by key.
functimm.models._registry.is_model(model_name:str) -> bool
Check if a model name exists
functimm.models._registry.model_entrypoint(model_name:str, module_filter:Optional[str]=None) -> Callable[..., Any]
Fetch a model entrypoint for specified model name
methodtimm.models.beit.Attention.forward(x:torch.Tensor, shared_rel_pos_bias:Optional[torch.Tensor]=None) -> torch.Tensor
Forward pass of attention module.
methodtimm.models.beit.Attention.init_non_persistent_buffers() -> None
Initialize non-persistent buffers.
methodtimm.models.beit.Attention.reset_parameters() -> None
Initialize parameters and buffers.
classtimm.models.beit.Beit
BEiT: BERT Pre-Training of Image Transformers.
methodtimm.models.beit.Beit.fix_init_weight() -> None
Fix initialization weights according to BEiT paper.
methodtimm.models.beit.Beit.forward(x:torch.Tensor) -> torch.Tensor
Forward pass through the model.
methodtimm.models.beit.Beit.forward_features(x:torch.Tensor) -> torch.Tensor
Forward pass through feature extraction layers.
methodtimm.models.beit.Beit.forward_head(x:torch.Tensor, pre_logits:bool=False) -> torch.Tensor
Forward pass through classification head.
methodtimm.models.beit.Beit.get_classifier() -> nn.Module
Get the classifier head.
methodtimm.models.beit.Beit.init_weights(needs_reset:bool=True) -> None
Initialize model weights.
methodtimm.models.beit.Beit.reset_classifier(num_classes:int, global_pool:Optional[str]=None)
Reset the classification head.
methodtimm.models.beit.Beit.set_grad_checkpointing(enable:bool=True)
Enable or disable gradient checkpointing.
classtimm.models.beit.Block
Transformer block with attention and MLP.
methodtimm.models.beit.Block.forward(x:torch.Tensor, shared_rel_pos_bias:Optional[torch.Tensor]=None) -> torch.Tensor
Forward pass of transformer block.
methodtimm.models.beit.Block.reset_parameters() -> None
Initialize parameters.
methodtimm.models.beit.RelativePositionBias.forward() -> torch.Tensor
Generate relative position bias.
methodtimm.models.beit.RelativePositionBias.init_non_persistent_buffers() -> None
Initialize non-persistent buffers.
methodtimm.models.beit.RelativePositionBias.reset_parameters() -> None
Initialize parameters and buffers.
functimm.models.beit.beit_base_patch16_224(pretrained:bool=False, **kwargs) -> Beit
BEiT base model @ 224x224 with patch size 16x16.
functimm.models.beit.beit_base_patch16_384(pretrained:bool=False, **kwargs) -> Beit
BEiT base model @ 384x384 with patch size 16x16.
functimm.models.beit.beit_large_patch16_224(pretrained:bool=False, **kwargs) -> Beit
BEiT large model @ 224x224 with patch size 16x16.
functimm.models.beit.beit_large_patch16_384(pretrained:bool=False, **kwargs) -> Beit
BEiT large model @ 384x384 with patch size 16x16.
functimm.models.beit.beit_large_patch16_512(pretrained:bool=False, **kwargs) -> Beit
BEiT large model @ 512x512 with patch size 16x16.
functimm.models.beit.beitv2_base_patch16_224(pretrained:bool=False, **kwargs) -> Beit
BEiT v2 base model @ 224x224 with patch size 16x16.
functimm.models.beit.beitv2_large_patch16_224(pretrained:bool=False, **kwargs) -> Beit
BEiT v2 large model @ 224x224 with patch size 16x16.
functimm.models.byoanet.botnet26t_256(pretrained:bool=False, **kwargs) -> ByobNet
Bottleneck Transformer w/ ResNet26-T backbone.
functimm.models.byoanet.botnet50ts_256(pretrained:bool=False, **kwargs) -> ByobNet
Bottleneck Transformer w/ ResNet50-T backbone, silu act.
functimm.models.byoanet.eca_botnext26ts_256(pretrained:bool=False, **kwargs) -> ByobNet
Bottleneck Transformer w/ ResNet26-T backbone, silu act.
functimm.models.byoanet.eca_halonext26ts(pretrained:bool=False, **kwargs) -> ByobNet
HaloNet w/ a ResNet26-t backbone, silu act.
functimm.models.byoanet.halo2botnet50ts_256(pretrained:bool=False, **kwargs) -> ByobNet
Combo Attention (Halo + Halo + Bot) Network
functimm.models.byoanet.halonet26t(pretrained:bool=False, **kwargs) -> ByobNet
HaloNet w/ a ResNet26-t backbone.
functimm.models.byoanet.halonet50ts(pretrained:bool=False, **kwargs) -> ByobNet
HaloNet w/ a ResNet50-t backbone, silu act.
functimm.models.byoanet.halonet_h1(pretrained:bool=False, **kwargs) -> ByobNet
HaloNet-H1.
functimm.models.byoanet.haloregnetz_b(pretrained:bool=False, **kwargs) -> ByobNet
Halo + RegNetZ
functimm.models.byoanet.lambda_resnet26rpt_256(pretrained:bool=False, **kwargs) -> ByobNet
Lambda-ResNet-26-R-T.
functimm.models.byoanet.lambda_resnet26t(pretrained:bool=False, **kwargs) -> ByobNet
Lambda-ResNet-26-T.
functimm.models.byoanet.lambda_resnet50ts(pretrained:bool=False, **kwargs) -> ByobNet
Lambda-ResNet-50-TS.
functimm.models.byoanet.lamhalobotnet50ts_256(pretrained:bool=False, **kwargs) -> ByobNet
Combo Attention (Lambda + Halo + Bot) Network
classtimm.models.byobnet.BasicBlock
ResNet Basic Block - kxk + kxk
classtimm.models.byobnet.BottleneckBlock
ResNet-like Bottleneck Block - 1x1 - kxk - 1x1
classtimm.models.byobnet.ByoBlockCfg
Block configuration for Bring-Your-Own-Blocks.
classtimm.models.byobnet.ByoModelCfg
Model configuration for Bring-Your-Own-Blocks network.
classtimm.models.byobnet.ByobNet
Bring-your-own-blocks Network.
methodtimm.models.byobnet.ByobNet.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
methodtimm.models.byobnet.ByobNet.forward_features(x:torch.Tensor) -> torch.Tensor
Forward pass through feature extraction.
methodtimm.models.byobnet.ByobNet.forward_head(x:torch.Tensor, pre_logits:bool=False) -> torch.Tensor
Forward pass through head.
methodtimm.models.byobnet.ByobNet.get_classifier() -> nn.Module
Get classifier module.
methodtimm.models.byobnet.ByobNet.group_matcher(coarse:bool=False) -> Dict[str, Any]
Group matcher for parameter groups.
methodtimm.models.byobnet.ByobNet.reset_classifier(num_classes:int, global_pool:Optional[str]=None) -> None
Reset classifier.
methodtimm.models.byobnet.ByobNet.set_grad_checkpointing(enable:bool=True) -> None
Enable or disable gradient checkpointing.
classtimm.models.byobnet.DownsampleAvg
Average pool downsampling module.
methodtimm.models.byobnet.DownsampleAvg.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
classtimm.models.byobnet.LayerFn
Container for layer factory functions.
classtimm.models.byobnet.MobileOneBlock
MobileOne building block.
methodtimm.models.byobnet.MobileOneBlock.forward(x:torch.Tensor) -> torch.Tensor
Apply forward pass.
classtimm.models.byobnet.RepVggBlock
RepVGG Block.
functimm.models.byobnet.drop_blocks(drop_prob:float=0.0, block_size:int=3, num_stages:int=4) -> List[Optional[partial]]
Create DropBlock layer partials for each stage.
functimm.models.byobnet.interleave_blocks(types:Tuple[str, str], d:int, every:Union[int, List[int]]=1, first:bool=False, **kwargs) -> Tuple[ByoBlockCfg, ...]
Interleave 2 block types in stack.
functimm.models.byobnet.num_groups(group_size:Optional[int], channels:int) -> int
Calculate number of groups for grouped convolution.
functimm.models.byobnet.resnet101_clip(pretrained=False, **kwargs) -> ByobNet
OpenAI Modified ResNet-101 CLIP image tower
functimm.models.byobnet.resnet50_clip(pretrained=False, **kwargs) -> ByobNet
OpenAI Modified ResNet-50 CLIP image tower
functimm.models.byobnet.resnet50x16_clip(pretrained=False, **kwargs) -> ByobNet
OpenAI Modified ResNet-50x16 CLIP image tower
functimm.models.byobnet.resnet50x4_clip(pretrained=False, **kwargs) -> ByobNet
OpenAI Modified ResNet-50x4 CLIP image tower
functimm.models.byobnet.resnet50x64_clip(pretrained=False, **kwargs) -> ByobNet
OpenAI Modified ResNet-50x64 CLIP image tower
classtimm.models.coat.CoaT
CoaT class.
classtimm.models.coat.ConvPosEnc
Convolutional Position Encoding.
classtimm.models.coat.ConvRelPosEnc
Convolutional relative position encoding.
classtimm.models.coat.ParallelBlock
Parallel block class.
methodtimm.models.coat.ParallelBlock.downsample(x, factor:float, size:Tuple[int, int])
Feature map down-sampling.
methodtimm.models.coat.ParallelBlock.interpolate(x, scale_factor:float, size:Tuple[int, int])
Feature map interpolation.
methodtimm.models.coat.ParallelBlock.upsample(x, factor:float, size:Tuple[int, int])
Feature map up-sampling.
classtimm.models.coat.SerialBlock
Serial block class.
functimm.models.coat.insert_cls(x, cls_token)
Insert CLS token.
functimm.models.coat.remove_cls(x)
Remove CLS token.
classtimm.models.convnext.ConvNeXt
ConvNeXt model architecture.
methodtimm.models.convnext.ConvNeXt.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
methodtimm.models.convnext.ConvNeXt.forward_features(x:torch.Tensor) -> torch.Tensor
Forward pass through feature extraction layers.
methodtimm.models.convnext.ConvNeXt.forward_head(x:torch.Tensor, pre_logits:bool=False) -> torch.Tensor
Forward pass through classifier head.
methodtimm.models.convnext.ConvNeXt.get_classifier() -> nn.Module
Get the classifier module.
methodtimm.models.convnext.ConvNeXt.group_matcher(coarse:bool=False) -> Dict[str, Union[str, List]]
Create regex patterns for parameter grouping.
methodtimm.models.convnext.ConvNeXt.reset_classifier(num_classes:int, global_pool:Optional[str]=None) -> None
Reset the classifier head.
methodtimm.models.convnext.ConvNeXt.set_grad_checkpointing(enable:bool=True) -> None
Enable or disable gradient checkpointing.
classtimm.models.convnext.ConvNeXtBlock
ConvNeXt Block.
methodtimm.models.convnext.ConvNeXtBlock.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
classtimm.models.convnext.ConvNeXtStage
ConvNeXt stage (multiple blocks).
methodtimm.models.convnext.ConvNeXtStage.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
classtimm.models.convnext.Downsample
Downsample module for ConvNeXt.
methodtimm.models.convnext.Downsample.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
functimm.models.convnext.checkpoint_filter_fn(state_dict, model)
Remap FB checkpoints -> timm
classtimm.models.crossvit.PatchEmbed
Image to Patch Embedding
classtimm.models.csatv2.Block
ConvNeXt-style block with spatial attention.
classtimm.models.csatv2.Dct1d
1D Discrete Cosine Transform layer.
classtimm.models.csatv2.Dct2d
2D Discrete Cosine Transform layer.
classtimm.models.csatv2.Dct2dStats
Utility module to compute DCT coefficient statistics.
classtimm.models.csatv2.PosConv
Convolutional position encoding.
functimm.models.csatv2.checkpoint_filter_fn(state_dict:dict, model:nn.Module) -> dict
Remap original CSATv2 checkpoint to timm format.
classtimm.models.cspnet.BottleneckBlock
ResNe(X)t Bottleneck Block
classtimm.models.cspnet.CrossStage
Cross Stage.
classtimm.models.cspnet.CrossStage3
Cross Stage 3.
classtimm.models.cspnet.CspNet
Cross Stage Partial base model.
classtimm.models.cspnet.DarkBlock
DarkNet Block
classtimm.models.cspnet.DarkStage
DarkNet stage.
classtimm.models.davit.SpatialBlock
Windows Block.
classtimm.models.densenet.DenseBlock
DenseNet Block.
methodtimm.models.densenet.DenseBlock.forward(init_features:torch.Tensor) -> torch.Tensor
Forward pass through all layers in the block.
classtimm.models.densenet.DenseLayer
Dense layer for DenseNet.
methodtimm.models.densenet.DenseLayer.any_requires_grad(x:List[torch.Tensor]) -> bool
Check if any tensor in list requires gradient.
methodtimm.models.densenet.DenseLayer.bottleneck_fn(xs:List[torch.Tensor]) -> torch.Tensor
Bottleneck function for concatenated features.
methodtimm.models.densenet.DenseLayer.call_checkpoint_bottleneck(x:List[torch.Tensor]) -> torch.Tensor
Call bottleneck function with gradient checkpointing.
methodtimm.models.densenet.DenseLayer.forward(x:Union[torch.Tensor, List[torch.Tensor]]) -> torch.Tensor
Forward pass.
classtimm.models.densenet.DenseNet
Densenet-BC model class.
methodtimm.models.densenet.DenseNet.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
methodtimm.models.densenet.DenseNet.forward_features(x:torch.Tensor) -> torch.Tensor
Forward pass through feature extraction layers.
methodtimm.models.densenet.DenseNet.forward_head(x:torch.Tensor, pre_logits:bool=False) -> torch.Tensor
Forward pass through classifier head.
methodtimm.models.densenet.DenseNet.get_classifier() -> nn.Module
Get the classifier head.
methodtimm.models.densenet.DenseNet.group_matcher(coarse:bool=False) -> Dict[str, Any]
Group parameters for optimization.
methodtimm.models.densenet.DenseNet.reset_classifier(num_classes:int, global_pool:str='avg') -> None
Reset the classifier head.
methodtimm.models.densenet.DenseNet.set_grad_checkpointing(enable:bool=True) -> None
Enable or disable gradient checkpointing.
classtimm.models.densenet.DenseTransition
Transition layer between DenseNet blocks.
classtimm.models.dla.DlaBasic
DLA Basic
classtimm.models.dla.DlaBottleneck
DLA/DLA-X Bottleneck
functimm.models.edgenext.checkpoint_filter_fn(state_dict, model)
Remap FB checkpoints -> timm
classtimm.models.efficientformer.ConvMlpWithNorm
Implementation of MLP with 1*1 convolutions.
methodtimm.models.efficientformer_v2.Attention2d.init_non_persistent_buffers() -> None
Initialize non-persistent buffers.
methodtimm.models.efficientformer_v2.Attention2d.reset_parameters() -> None
Initialize parameters and buffers.
classtimm.models.efficientformer_v2.ConvMlpWithNorm
Implementation of MLP with 1*1 convolutions.
classtimm.models.efficientnet.EfficientNet
EfficientNet model architecture.
methodtimm.models.efficientnet.EfficientNet.forward(x:torch.Tensor) -> torch.Tensor
Forward pass.
methodtimm.models.efficientnet.EfficientNet.forward_features(x:torch.Tensor) -> torch.Tensor
Forward pass through feature extraction layers.
methodtimm.models.efficientnet.EfficientNet.forward_head(x:torch.Tensor, pre_logits:bool=False) -> torch.Tensor
Forward pass through classifier head.
methodtimm.models.efficientnet.EfficientNet.get_classifier() -> nn.Module
Get the classifier module.
methodtimm.models.efficientnet.EfficientNet.group_matcher(coarse:bool=False) -> Dict[str, Union[str, List]]
Create regex patterns for parameter groups.
methodtimm.models.efficientnet.EfficientNet.reset_classifier(num_classes:int, global_pool:str='avg') -> None
Reset the classifier head.
methodtimm.models.efficientnet.EfficientNet.set_grad_checkpointing(enable:bool=True) -> None
Enable or disable gradient checkpointing.
functimm.models.efficientnet.efficientnet_b0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B0
functimm.models.efficientnet.efficientnet_b0_g8_gn(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B0 w/ group conv + GroupNorm
functimm.models.efficientnet.efficientnet_b0_gn(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B0 + GroupNorm
functimm.models.efficientnet.efficientnet_b1(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B1
functimm.models.efficientnet.efficientnet_b1_pruned(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B1 Pruned.
functimm.models.efficientnet.efficientnet_b2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B2
functimm.models.efficientnet.efficientnet_b2_pruned(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B2 Pruned.
functimm.models.efficientnet.efficientnet_b3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3
functimm.models.efficientnet.efficientnet_b3_g8_gn(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3 w/ grouped conv + BN
functimm.models.efficientnet.efficientnet_b3_gn(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3 w/ GroupNorm
functimm.models.efficientnet.efficientnet_b3_pruned(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3 Pruned.
functimm.models.efficientnet.efficientnet_b4(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B4
functimm.models.efficientnet.efficientnet_b5(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B5
functimm.models.efficientnet.efficientnet_b6(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B6
functimm.models.efficientnet.efficientnet_b7(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B7
functimm.models.efficientnet.efficientnet_b8(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B8
functimm.models.efficientnet.efficientnet_blur_b0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B0 w/ BlurPool
functimm.models.efficientnet.efficientnet_cc_b0_4e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B0 w/ 8 Experts
functimm.models.efficientnet.efficientnet_cc_b0_8e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B0 w/ 8 Experts
functimm.models.efficientnet.efficientnet_cc_b1_8e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B1 w/ 8 Experts
functimm.models.efficientnet.efficientnet_el(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge-Large.
functimm.models.efficientnet.efficientnet_el_pruned(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge-Large pruned.
functimm.models.efficientnet.efficientnet_em(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge-Medium.
functimm.models.efficientnet.efficientnet_es(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge Small.
functimm.models.efficientnet.efficientnet_es_pruned(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge Small Pruned.
functimm.models.efficientnet.efficientnet_h_b5(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B5
functimm.models.efficientnet.efficientnet_l2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-L2.
functimm.models.efficientnet.efficientnet_lite0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite0
functimm.models.efficientnet.efficientnet_lite1(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite1
functimm.models.efficientnet.efficientnet_lite2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite2
functimm.models.efficientnet.efficientnet_lite3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite3
functimm.models.efficientnet.efficientnet_lite4(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite4
functimm.models.efficientnet.efficientnet_x_b3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3
functimm.models.efficientnet.efficientnet_x_b5(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B5
functimm.models.efficientnet.efficientnetv2_l(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Large.
functimm.models.efficientnet.efficientnetv2_m(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Medium.
functimm.models.efficientnet.efficientnetv2_rw_m(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Medium (RW variant).
functimm.models.efficientnet.efficientnetv2_rw_s(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Small (RW variant).
functimm.models.efficientnet.efficientnetv2_s(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Small.
functimm.models.efficientnet.efficientnetv2_xl(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Xtra-Large.
functimm.models.efficientnet.fbnetc_100(pretrained=False, **kwargs) -> EfficientNet
FBNet-C
functimm.models.efficientnet.mixnet_l(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Large model.
functimm.models.efficientnet.mixnet_m(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Medium model.
functimm.models.efficientnet.mixnet_s(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Small model.
functimm.models.efficientnet.mixnet_xl(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Extra-Large model.
functimm.models.efficientnet.mixnet_xxl(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Double Extra Large model.
functimm.models.efficientnet.mnasnet_050(pretrained=False, **kwargs) -> EfficientNet
MNASNet B1, depth multiplier of 0.5.
functimm.models.efficientnet.mnasnet_075(pretrained=False, **kwargs) -> EfficientNet
MNASNet B1, depth multiplier of 0.75.
functimm.models.efficientnet.mnasnet_100(pretrained=False, **kwargs) -> EfficientNet
MNASNet B1, depth multiplier of 1.0.
functimm.models.efficientnet.mnasnet_140(pretrained=False, **kwargs) -> EfficientNet
MNASNet B1, depth multiplier of 1.4
functimm.models.efficientnet.mnasnet_small(pretrained=False, **kwargs) -> EfficientNet
MNASNet Small, depth multiplier of 1.0.
functimm.models.efficientnet.mobilenet_edgetpu_100(pretrained=False, **kwargs) -> EfficientNet
MobileNet-EdgeTPU-v1 100.
functimm.models.efficientnet.mobilenet_edgetpu_v2_l(pretrained=False, **kwargs) -> EfficientNet
MobileNet-EdgeTPU-v2 Large.
functimm.models.efficientnet.mobilenet_edgetpu_v2_m(pretrained=False, **kwargs) -> EfficientNet
MobileNet-EdgeTPU-v2 Medium.
functimm.models.efficientnet.mobilenet_edgetpu_v2_s(pretrained=False, **kwargs) -> EfficientNet
MobileNet-EdgeTPU-v2 Small.
functimm.models.efficientnet.mobilenet_edgetpu_v2_xs(pretrained=False, **kwargs) -> EfficientNet
MobileNet-EdgeTPU-v2 Extra Small.
functimm.models.efficientnet.mobilenetv1_100(pretrained=False, **kwargs) -> EfficientNet
MobileNet V1
functimm.models.efficientnet.mobilenetv1_100h(pretrained=False, **kwargs) -> EfficientNet
MobileNet V1
functimm.models.efficientnet.mobilenetv1_125(pretrained=False, **kwargs) -> EfficientNet
MobileNet V1
functimm.models.efficientnet.mobilenetv2_035(pretrained=False, **kwargs) -> EfficientNet
MobileNet V2 w/ 0.35 channel multiplier
functimm.models.efficientnet.mobilenetv2_050(pretrained=False, **kwargs) -> EfficientNet
MobileNet V2 w/ 0.5 channel multiplier
functimm.models.efficientnet.mobilenetv2_075(pretrained=False, **kwargs) -> EfficientNet
MobileNet V2 w/ 0.75 channel multiplier
functimm.models.efficientnet.mobilenetv2_100(pretrained=False, **kwargs) -> EfficientNet
MobileNet V2 w/ 1.0 channel multiplier
functimm.models.efficientnet.mobilenetv2_140(pretrained=False, **kwargs) -> EfficientNet
MobileNet V2 w/ 1.4 channel multiplier
functimm.models.efficientnet.semnasnet_050(pretrained=False, **kwargs) -> EfficientNet
MNASNet A1 (w/ SE), depth multiplier of 0.5
functimm.models.efficientnet.semnasnet_100(pretrained=False, **kwargs) -> EfficientNet
MNASNet A1 (w/ SE), depth multiplier of 1.0.
functimm.models.efficientnet.semnasnet_140(pretrained=False, **kwargs) -> EfficientNet
MNASNet A1 (w/ SE), depth multiplier of 1.4.
functimm.models.efficientnet.spnasnet_100(pretrained=False, **kwargs) -> EfficientNet
Single-Path NAS Pixel1
functimm.models.efficientnet.tf_efficientnet_b0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B0.
functimm.models.efficientnet.tf_efficientnet_b1(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B1.
functimm.models.efficientnet.tf_efficientnet_b2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B2.
functimm.models.efficientnet.tf_efficientnet_b3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B3.
functimm.models.efficientnet.tf_efficientnet_b4(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B4.
functimm.models.efficientnet.tf_efficientnet_b5(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B5.
functimm.models.efficientnet.tf_efficientnet_b6(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B6.
functimm.models.efficientnet.tf_efficientnet_b7(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B7.
functimm.models.efficientnet.tf_efficientnet_b8(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-B8.
functimm.models.efficientnet.tf_efficientnet_cc_b0_4e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B0 w/ 4 Experts.
functimm.models.efficientnet.tf_efficientnet_cc_b0_8e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B0 w/ 8 Experts.
functimm.models.efficientnet.tf_efficientnet_cc_b1_8e(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-CondConv-B1 w/ 8 Experts.
functimm.models.efficientnet.tf_efficientnet_el(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge-Large.
functimm.models.efficientnet.tf_efficientnet_em(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge-Medium.
functimm.models.efficientnet.tf_efficientnet_es(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Edge Small.
functimm.models.efficientnet.tf_efficientnet_l2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-L2 NoisyStudent.
functimm.models.efficientnet.tf_efficientnet_lite0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite0
functimm.models.efficientnet.tf_efficientnet_lite1(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite1
functimm.models.efficientnet.tf_efficientnet_lite2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite2
functimm.models.efficientnet.tf_efficientnet_lite3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite3
functimm.models.efficientnet.tf_efficientnet_lite4(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-Lite4
functimm.models.efficientnet.tf_efficientnetv2_b0(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2-B0.
functimm.models.efficientnet.tf_efficientnetv2_b1(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2-B1.
functimm.models.efficientnet.tf_efficientnetv2_b2(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2-B2.
functimm.models.efficientnet.tf_efficientnetv2_b3(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2-B3.
functimm.models.efficientnet.tf_efficientnetv2_l(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Large.
functimm.models.efficientnet.tf_efficientnetv2_m(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Medium.
functimm.models.efficientnet.tf_efficientnetv2_s(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Small.
functimm.models.efficientnet.tf_efficientnetv2_xl(pretrained=False, **kwargs) -> EfficientNet
EfficientNet-V2 Xtra-Large.
functimm.models.efficientnet.tf_mixnet_l(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Large model.
functimm.models.efficientnet.tf_mixnet_m(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Medium model.
functimm.models.efficientnet.tf_mixnet_s(pretrained=False, **kwargs) -> EfficientNet
Creates a MixNet Small model.
classtimm.models.efficientvit_mit.LiteMLA
Lightweight multi-scale linear attention
classtimm.models.efficientvit_msra.EfficientVitBlock
A basic EfficientVit building block.
classtimm.models.efficientvit_msra.LocalWindowAttention
Local Window Attention.
functimm.models.eva.eva02_base_patch14_224(pretrained:bool=False, **kwargs) -> Eva
EVA02 Base https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_base_patch14_448(pretrained:bool=False, **kwargs) -> Eva
EVA02 Base https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_large_patch14_224(pretrained:bool=False, **kwargs) -> Eva
EVA02 Large https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_large_patch14_448(pretrained:bool=False, **kwargs) -> Eva
EVA02 Large https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_small_patch14_224(pretrained:bool=False, **kwargs) -> Eva
EVA02 Small https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_small_patch14_336(pretrained:bool=False, **kwargs) -> Eva
EVA02 Small https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_tiny_patch14_224(pretrained:bool=False, **kwargs) -> Eva
EVA02 Tiny https://arxiv.org/abs/2303.11331
functimm.models.eva.eva02_tiny_patch14_336(pretrained:bool=False, **kwargs) -> Eva
EVA02 Tiny https://arxiv.org/abs/2303.11331
functimm.models.eva.eva_giant_patch14_224(pretrained:bool=False, **kwargs) -> Eva
EVA-g model https://arxiv.org/abs/2211.07636
functimm.models.eva.eva_giant_patch14_336(pretrained:bool=False, **kwargs) -> Eva
EVA-g model https://arxiv.org/abs/2211.07636
functimm.models.eva.eva_giant_patch14_560(pretrained:bool=False, **kwargs) -> Eva
EVA-g model https://arxiv.org/abs/2211.07636
functimm.models.eva.vit_base_patch16_rope_reg1_gap_256(pretrained:bool=False, **kwargs) -> Eva
timm SBB ViT with ROPE
functimm.models.eva.vit_betwixt_patch16_rope_reg4_gap_256(pretrained:bool=False, **kwargs) -> Eva
timm SBB ViT with ROPE
functimm.models.eva.vit_medium_patch16_rope_reg1_gap_256(pretrained:bool=False, **kwargs) -> Eva
timm SBB ViT with ROPE
functimm.models.eva.vit_mediumd_patch16_rope_reg1_gap_256(pretrained:bool=False, **kwargs) -> Eva
timm SBB ViT with ROPE
functimm.models.eva.vit_tiny_patch16_dinov3_qkvb(pretrained:bool=False, **kwargs) -> Eva
DINOv3-style T/16 w/ QKV bias enabled.
classtimm.models.fastvit.Attention
Multi-headed Self Attention module.
classtimm.models.fastvit.ConvMlp
Convolutional FFN Module.
classtimm.models.fastvit.MobileOneBlock
MobileOne building block.
methodtimm.models.fastvit.MobileOneBlock.forward(x:torch.Tensor) -> torch.Tensor
Apply forward pass.
classtimm.models.fastvit.PatchEmbed
Convolutional patch embedding layer.
classtimm.models.fastvit.RepConditionalPosEnc
Implementation of conditional positional encoding.
classtimm.models.fastvit.RepMixer
Reparameterizable token mixer.
functimm.models.fastvit.fastvit_mci3(pretrained=False, **kwargs)
Instantiate L model variant.
functimm.models.fastvit.fastvit_mci4(pretrained=False, **kwargs)
Instantiate XL model variant.
classtimm.models.focalnet.FocalNet
" Focal Modulation Networks (FocalNets)
classtimm.models.focalnet.FocalNetBlock
Focal Modulation Network Block.
classtimm.models.focalnet.FocalNetStage
A basic Focal Transformer layer for one stage.
classtimm.models.gemma4_vit.Gemma4ClippableLinear
Linear layer with optional input/output clamping.
classtimm.models.gemma4_vit.Gemma4GatedMlp
Gated MLP for Gemma4 Vision Encoder.
classtimm.models.gemma4_vit.Gemma4VisionPooler
Spatial pooling for Gemma4 vision encoder output.
classtimm.models.gemma4_vit.Gemma4VitClassifier
Classification wrapper around ``Gemma4VitEncoder``.
methodtimm.models.gemma4_vit.Gemma4VitClassifier.forward_features(x:Union[torch.Tensor, Dict[str, torch.Tensor]], patch_coord:Optional[torch.Tensor]=None, patch_valid:Optional[torch.Tensor]=None) -> torch.Tensor
Pre-head features.
methodtimm.models.gemma4_vit.Gemma4VitClassifier.get_patch_size() -> Tuple[int, int]
Return the 2-tuple patch size.
classtimm.models.gemma4_vit.Gemma4VitEncoder
Gemma4 Vision Encoder.
methodtimm.models.gemma4_vit.Gemma4VitEncoder.forward_features(x:Union[torch.Tensor, Dict[str, torch.Tensor]], patch_coord:Optional[torch.Tensor]=None, patch_valid:Optional[torch.Tensor]=None) -> torch.Tensor
Raw patch tokens pre-pool.
methodtimm.models.gemma4_vit.Gemma4VitEncoder.get_patch_size() -> Tuple[int, int]
Return the 2-tuple patch size.
methodtimm.models.gemma4_vit.Gemma4VitEncoder.init_weights(mode:str='', needs_reset:bool=True) -> None
Initialize model weights.
methodtimm.models.gemma4_vit.Gemma4VitEncoder.set_clamp_enabled(enabled:bool=True) -> None
Toggle the ``Gemma4ClippableLinear`` clamp ops.
functimm.models.gemma4_vit.apply_rotary_pos_emb(x:torch.Tensor, cos:torch.Tensor, sin:torch.Tensor, unsqueeze_dim:int=2) -> torch.Tensor
Apply rotary position embedding to input tensor.
functimm.models.gemma4_vit.gemma4_vit_167m(pretrained:bool=False, **kwargs) -> Gemma4VitClassifier
Gemma4 ~167M (E2B/E4B vision tower) classifier.
functimm.models.gemma4_vit.gemma4_vit_570m(pretrained:bool=False, **kwargs) -> Gemma4VitClassifier
Gemma4 ~570M (26B/31B vision tower) classifier.
functimm.models.gemma4_vit.rotate_half(x:torch.Tensor) -> torch.Tensor
Rotates half the hidden dims of the input.
classtimm.models.ghostnet.GhostBottleneck
GhostV1/V2 bottleneck w/ optional SE
classtimm.models.ghostnet.GhostBottleneckV3
GhostV3 bottleneck w/ optional SE
functimm.models.ghostnet.ghostnet_050(pretrained=False, **kwargs) -> GhostNet
GhostNet-0.5x
functimm.models.ghostnet.ghostnet_100(pretrained=False, **kwargs) -> GhostNet
GhostNet-1.0x
functimm.models.ghostnet.ghostnet_130(pretrained=False, **kwargs) -> GhostNet
GhostNet-1.3x
functimm.models.ghostnet.ghostnetv2_100(pretrained=False, **kwargs) -> GhostNet
GhostNetV2-1.0x
functimm.models.ghostnet.ghostnetv2_130(pretrained=False, **kwargs) -> GhostNet
GhostNetV2-1.3x
functimm.models.ghostnet.ghostnetv2_160(pretrained=False, **kwargs) -> GhostNet
GhostNetV2-1.6x
functimm.models.ghostnet.ghostnetv3_050(pretrained:bool=False, **kwargs:Any) -> GhostNet
GhostNetV3-0.5x
functimm.models.ghostnet.ghostnetv3_100(pretrained:bool=False, **kwargs:Any) -> GhostNet
GhostNetV3-1.0x
functimm.models.ghostnet.ghostnetv3_130(pretrained:bool=False, **kwargs:Any) -> GhostNet
GhostNetV3-1.3x
functimm.models.ghostnet.ghostnetv3_160(pretrained:bool=False, **kwargs:Any) -> GhostNet
GhostNetV3-1.6x
functimm.models.hardcorenas.hardcorenas_a(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_A
functimm.models.hardcorenas.hardcorenas_b(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_B
functimm.models.hardcorenas.hardcorenas_c(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_C
functimm.models.hardcorenas.hardcorenas_d(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_D
functimm.models.hardcorenas.hardcorenas_e(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_E
functimm.models.hardcorenas.hardcorenas_f(pretrained=False, **kwargs) -> MobileNetV3
hardcorenas_F
classtimm.models.hiera.MaskUnitAttention
Computes either Mask Unit or Global Attention.
methodtimm.models.hiera.MaskUnitAttention.forward(x:torch.Tensor) -> torch.Tensor
Input should be of shape [batch, tokens, channels].
functimm.models.hiera.conv_nd(n:int) -> Type[nn.Module]
Returns a conv with nd (e.g., Conv2d for n=2).
classtimm.models.hieradet_sam2.HieraDet
Reference: https://arxiv.org/abs/2306.00989
classtimm.models.hieradet_sam2.HieraPatchEmbed
Image to Patch Embedding.
classtimm.models.inception_next.InceptionDWConv2d
Inception depthwise convolution
classtimm.models.inception_next.MlpClassifierHead
MLP classification head
classtimm.models.inception_v3.InceptionV3
Inception-V3
methodtimm.models.levit.Attention.init_non_persistent_buffers() -> None
Initialize non-persistent buffers.
methodtimm.models.levit.Attention.reset_parameters() -> None
Initialize parameters and buffers.
classtimm.models.mambaout.MlpHead
MLP classification head
classtimm.models.maxxvit.Attention2d
Multi-head attention for 2D NCHW tensors.
classtimm.models.maxxvit.AttentionCl
Channels-last multi-head attention (B, ..., C).
classtimm.models.maxxvit.ConvNeXtBlock
ConvNeXt Block.
classtimm.models.maxxvit.MaxxVit
CoaTNet + MaxVit base model.
classtimm.models.maxxvit.MaxxVitCfg
Configuration for MaxxVit models.
classtimm.models.maxxvit.MaxxVitConvCfg
Configuration for MaxxVit convolution blocks.
classtimm.models.maxxvit.MaxxVitTransformerCfg
Configuration for MaxxVit transformer blocks.
classtimm.models.maxxvit.ParallelPartitionAttention
Experimental.
classtimm.models.maxxvit.PartitionAttention2d
Grid or Block partition + Attn + FFN.
classtimm.models.maxxvit.PartitionAttentionCl
Grid or Block partition + Attn + FFN.
classtimm.models.maxxvit.Stem
Stem layer for feature extraction.
classtimm.models.maxxvit.TransformerBlock2d
Transformer block with 2D downsampling.
functimm.models.maxxvit.checkpoint_filter_fn(state_dict:Dict[str, torch.Tensor], model:nn.Module) -> Dict[str, torch.Tensor]
Filter checkpoint state dict for compatibility.
functimm.models.maxxvit.coatnet_0_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-0 model.
functimm.models.maxxvit.coatnet_0_rw_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-0 model with RW configuration.
functimm.models.maxxvit.coatnet_1_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-1 model.
functimm.models.maxxvit.coatnet_1_rw_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-1 model with RW configuration.
functimm.models.maxxvit.coatnet_2_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-2 model.
functimm.models.maxxvit.coatnet_2_rw_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-2 model with RW configuration.
functimm.models.maxxvit.coatnet_3_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-3 model.
functimm.models.maxxvit.coatnet_3_rw_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-3 model with RW configuration.
functimm.models.maxxvit.coatnet_4_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-4 model.
functimm.models.maxxvit.coatnet_5_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-5 model.
functimm.models.maxxvit.coatnet_bn_0_rw_224(pretrained:bool=False, **kwargs:Any) -> MaxxVit
CoatNet-0 model with BatchNorm and RW configuration.

About this data

These signatures were extracted from the public source of huggingface/pytorch-image-models using Python's ast module. Argument names, default values, type annotations and return types are taken verbatim from the code. Implementation bodies are never stored. See how it works for details.

Back to all 805 libraries