Add initial ConvNeXt impl (mods of official code)

Author: rwightman

timm/models/__init__.py

@@ -5,6 +5,7 @@
 from .coat import *
 from .convit import *
 from .convmixer import *
+from .convnext import *
 from .crossvit import *
 from .cspnet import *
 from .densenet import *

timm/models/convnext.py

@@ -0,0 +1,375 @@
+""" ConvNext
+
+Paper: `A ConvNet for the 2020s` - https://arxiv.org/pdf/2201.03545.pdf
+
+Original code and weights from https://github.com/facebookresearch/ConvNeXt, original copyright below
+
+Modifications and additions for timm hacked together by / Copyright 2022, Ross Wightman
+"""
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+# This source code is licensed under the MIT license
+
+from functools import partial
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .fx_features import register_notrace_module
+from .helpers import named_apply, build_model_with_cfg
+from .layers import trunc_normal_, ClassifierHead, SelectAdaptivePool2d, DropPath, ConvMlp, Mlp
+from .registry import register_model
+
+
+__all__ = ['ConvNeXt']  # model_registry will add each entrypoint fn to this
+
+
+def _cfg(url='', **kwargs):
+    return {
+        'url': url,
+        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
+        'crop_pct': 0.875, 'interpolation': 'bicubic',
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
+        'first_conv': 'stem.0', 'classifier': 'head',
+        **kwargs
+    }
+
+
+default_cfgs = dict(
+    convnext_tiny=_cfg(url="https://dl.fbaipublicfiles.com/convnext/convnext_tiny_1k_224_ema.pth"),
+    convnext_small=_cfg(url="https://dl.fbaipublicfiles.com/convnext/convnext_small_1k_224_ema.pth"),
+    convnext_base=_cfg(url="https://dl.fbaipublicfiles.com/convnext/convnext_base_1k_224_ema.pth"),
+    convnext_large=_cfg(url="https://dl.fbaipublicfiles.com/convnext/convnext_large_1k_224_ema.pth"),
+
+    convnext_tiny_hnf=_cfg(url='', classifier='head.fc'),
+
+    convnext_base_in22k=_cfg(
+        url="https://dl.fbaipublicfiles.com/convnext/convnext_base_22k_224.pth", num_classes=21841),
+    convnext_large_in22k=_cfg(
+        url="https://dl.fbaipublicfiles.com/convnext/convnext_large_22k_224.pth", num_classes=21841),
+    convnext_xlarge_in22k=_cfg(
+        url="https://dl.fbaipublicfiles.com/convnext/convnext_xlarge_22k_224.pth", num_classes=21841),
+)
+
+
+def _is_contiguous(tensor: torch.Tensor) -> bool:
+    # jit is oh so lovely :/
+    # if torch.jit.is_tracing():
+    #     return True
+    if torch.jit.is_scripting():
+        return tensor.is_contiguous()
+    else:
+        return tensor.is_contiguous(memory_format=torch.contiguous_format)
+
+
+@register_notrace_module
+class LayerNorm2d(nn.Module):
+    r""" LayerNorm for channels_first tensors with 2d spatial dimensions (ie N, C, H, W).
+    """
+
+    def __init__(self, normalized_shape, eps=1e-6):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(normalized_shape))
+        self.bias = nn.Parameter(torch.zeros(normalized_shape))
+        self.eps = eps
+        self.normalized_shape = (normalized_shape,)
+
+    def forward(self, x) -> torch.Tensor:
+        if _is_contiguous(x):
+            return F.layer_norm(
+                x.permute(0, 2, 3, 1), self.normalized_shape, self.weight, self.bias, self.eps).permute(0, 3, 1, 2)
+        else:
+            s, u = torch.var_mean(x, dim=1, keepdim=True)
+            x = (x - u) * torch.rsqrt(s + self.eps)
+            x = x * self.weight[:, None, None] + self.bias[:, None, None]
+            return x
+
+
+class ConvNeXtBlock(nn.Module):
+    """ ConvNeXt Block
+    There are two equivalent implementations:
+      (1) DwConv -> LayerNorm (channels_first) -> 1x1 Conv -> GELU -> 1x1 Conv; all in (N, C, H, W)
+      (2) DwConv -> Permute to (N, H, W, C); LayerNorm (channels_last) -> Linear -> GELU -> Linear; Permute back
+
+    Unlike the official impl, this one allows choice of 1 or 2, 1x1 conv can be faster with appropriate
+    choice of LayerNorm impl, however as model size increases the tradeoffs appear to change and nn.Linear
+    is a better choice. This was observed with PyTorch 1.10 on 3090 GPU, it could change over time & w/ different HW.
+
+    Args:
+        dim (int): Number of input channels.
+        drop_path (float): Stochastic depth rate. Default: 0.0
+        ls_init_value (float): Init value for Layer Scale. Default: 1e-6.
+    """
+
+    def __init__(self, dim, drop_path=0., ls_init_value=1e-6, conv_mlp=True, mlp_ratio=4, norm_layer=None):
+        super().__init__()
+        norm_layer = norm_layer or (partial(LayerNorm2d, eps=1e-6) if conv_mlp else partial(nn.LayerNorm, eps=1e-6))
+        mlp_layer = ConvMlp if conv_mlp else Mlp
+        self.use_conv_mlp = conv_mlp
+        self.conv_dw = nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim)  # depthwise conv
+        self.norm = norm_layer(dim)
+        self.mlp = mlp_layer(dim, int(mlp_ratio * dim), act_layer=nn.GELU)
+        self.gamma = nn.Parameter(ls_init_value * torch.ones(dim)) if ls_init_value > 0 else None
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+
+    def forward(self, x):
+        shortcut = x
+        x = self.conv_dw(x)
+        if self.use_conv_mlp:
+            x = self.norm(x)
+            x = self.mlp(x)
+            if self.gamma is not None:
+                x.mul_(self.gamma.reshape(1, -1, 1, 1))
+        else:
+            x = x.permute(0, 2, 3, 1)
+            x = self.norm(x)
+            x = self.mlp(x)
+            if self.gamma is not None:
+                x.mul_(self.gamma)
+            x = x.permute(0, 3, 1, 2)
+        x = self.drop_path(x) + shortcut
+        return x
+
+
+class ConvNeXtStage(nn.Module):
+
+    def __init__(
+            self, in_chs, out_chs, stride=2, depth=2, dp_rates=None, ls_init_value=1.0, conv_mlp=True,
+            norm_layer=None, cl_norm_layer=None, cross_stage=False):
+        super().__init__()
+
+        if in_chs != out_chs or stride > 1:
+            self.downsample = nn.Sequential(
+                norm_layer(in_chs),
+                nn.Conv2d(in_chs, out_chs, kernel_size=stride, stride=stride),
+            )
+        else:
+            self.downsample = nn.Identity()
+
+        dp_rates = dp_rates or [0.] * depth
+        self.blocks = nn.Sequential(*[ConvNeXtBlock(
+            dim=out_chs, drop_path=dp_rates[j], ls_init_value=ls_init_value, conv_mlp=conv_mlp,
+            norm_layer=norm_layer if conv_mlp else cl_norm_layer)
+            for j in range(depth)]
+        )
+
+    def forward(self, x):
+        x = self.downsample(x)
+        x = self.blocks(x)
+        return x
+
+
+class ConvNeXt(nn.Module):
+    r""" ConvNeXt
+        A PyTorch impl of : `A ConvNet for the 2020s`  - https://arxiv.org/pdf/2201.03545.pdf
+
+    Args:
+        in_chans (int): Number of input image channels. Default: 3
+        num_classes (int): Number of classes for classification head. Default: 1000
+        depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
+        dims (tuple(int)): Feature dimension at each stage. Default: [96, 192, 384, 768]
+        drop_rate (float): Head dropout rate
+        drop_path_rate (float): Stochastic depth rate. Default: 0.
+        ls_init_value (float): Init value for Layer Scale. Default: 1e-6.
+        head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
+    """
+
+    def __init__(
+            self, in_chans=3, num_classes=1000, global_pool='avg', output_stride=32, patch_size=4,
+            depths=(3, 3, 9, 3), dims=(96, 192, 384, 768),  ls_init_value=1e-6, conv_mlp=True,
+            head_init_scale=1., head_norm_first=False, norm_layer=None, drop_rate=0., drop_path_rate=0.,
+    ):
+        super().__init__()
+        assert output_stride == 32
+        if norm_layer is None:
+            norm_layer = partial(LayerNorm2d, eps=1e-6)
+            cl_norm_layer = norm_layer if conv_mlp else partial(nn.LayerNorm, eps=1e-6)
+        else:
+            assert conv_mlp,\
+                'If a norm_layer is specified, conv MLP must be used so all norm expect rank-4, channels-first input'
+            cl_norm_layer = norm_layer
+
+        partial(LayerNorm2d, eps=1e-6)
+        self.num_classes = num_classes
+        self.drop_rate = drop_rate
+        self.feature_info = []
+
+        # NOTE: this stem is a minimal form of ViT PatchEmbed, as used in SwinTransformer w/ patch_size = 4
+        self.stem = nn.Sequential(
+            nn.Conv2d(in_chans, dims[0], kernel_size=patch_size, stride=patch_size),
+            norm_layer(dims[0])
+        )
+
+        self.stages = nn.Sequential()
+        dp_rates = [x.tolist() for x in torch.linspace(0, drop_path_rate, sum(depths)).split(depths)]
+        curr_stride = patch_size
+        prev_chs = dims[0]
+        stages = []
+        # 4 feature resolution stages, each consisting of multiple residual blocks
+        for i in range(4):
+            stride = 2 if i > 0 else 1
+            # FIXME support dilation / output_stride
+            curr_stride *= stride
+            out_chs = dims[i]
+            stages.append(ConvNeXtStage(
+                prev_chs, out_chs, stride=stride,
+                depth=depths[i], dp_rates=dp_rates[i], ls_init_value=ls_init_value, conv_mlp=conv_mlp,
+                norm_layer=norm_layer, cl_norm_layer=cl_norm_layer)
+            )
+            prev_chs = out_chs
+            # NOTE feature_info use currently assumes stage 0 == stride 1, rest are stride 2
+            self.feature_info += [dict(num_chs=prev_chs, reduction=curr_stride, module=f'stages.{i}')]
+        self.stages = nn.Sequential(*stages)
+
+        self.num_features = prev_chs
+        if head_norm_first:
+            # norm -> global pool -> fc ordering, like most other nets (not compat with FB weights)
+            self.norm = norm_layer(self.num_features)  # final norm layer
+            self.pool = None  # global pool in ClassifierHead, pool == None being used to differentiate
+            self.head = ClassifierHead(self.num_features, num_classes, pool_type=global_pool, drop_rate=drop_rate)
+        else:
+            # pool -> norm -> fc, the default ConvNeXt ordering (pretrained FB weights)
+            self.pool = SelectAdaptivePool2d(pool_type=global_pool)
+            # NOTE when cl_norm_layer != norm_layer we could flatten here and use cl, but makes no performance diff
+            self.norm = norm_layer(self.num_features)
+            self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
+
+        named_apply(partial(_init_weights, head_init_scale=head_init_scale), self)
+
+    def get_classifier(self):
+        return self.head.fc if self.pool is None else self.head
+
+    def reset_classifier(self, num_classes=0, global_pool='avg'):
+        if self.pool is None:
+            # norm -> global pool -> fc ordering
+            self.head = ClassifierHead(
+                self.num_features, num_classes, pool_type=global_pool, drop_rate=self.drop_rate)
+        else:
+            # pool -> norm -> fc
+            self.pool = SelectAdaptivePool2d(pool_type=global_pool)
+            self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
+
+    def forward_features(self, x):
+        x = self.stem(x)
+        x = self.stages(x)
+        if self.pool is None:
+            # standard head, norm -> spatial pool -> fc
+            # ideally, last norm is within forward_features, but can only do so if norm precedes pooling
+            x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        if self.pool is not None:
+            # ConvNeXt head, spatial pool -> norm -> fc
+            # FIXME clean this up
+            x = self.pool(x)
+            x = self.norm(x)
+            if not self.pool.is_identity():
+                x = x.flatten(1)
+            if self.drop_rate > 0:
+                x = F.dropout(x, self.drop_rate, self.training)
+        x = self.head(x)
+        return x
+
+
+def _init_weights(module, name=None, head_init_scale=1.0):
+    if isinstance(module, nn.Conv2d):
+        trunc_normal_(module.weight, std=.02)
+        nn.init.constant_(module.bias, 0)
+    elif isinstance(module, nn.Linear):
+        trunc_normal_(module.weight, std=.02)
+        nn.init.constant_(module.bias, 0)
+        if name and '.head' in name:
+            module.weight.data.mul_(head_init_scale)
+            module.bias.data.mul_(head_init_scale)
+
+
+def checkpoint_filter_fn(state_dict, model):
+    """ Remap FB checkpoints -> timm """
+    if 'model' in state_dict:
+        state_dict = state_dict['model']
+    out_dict = {}
+    import re
+    for k, v in state_dict.items():
+        k = k.replace('downsample_layers.0.', 'stem.')
+        k = re.sub(r'stages.([0-9]+).([0-9]+)', r'stages.\1.blocks.\2', k)
+        k = re.sub(r'downsample_layers.([0-9]+).([0-9]+)', r'stages.\1.downsample.\2', k)
+        k = k.replace('dwconv', 'conv_dw')
+        k = k.replace('pwconv', 'mlp.fc')
+        if v.ndim == 2 and 'head' not in k:
+            model_shape = model.state_dict()[k].shape
+            v = v.reshape(model_shape)
+        out_dict[k] = v
+    return out_dict
+
+
+def _create_convnext(variant, pretrained=False, **kwargs):
+    model = build_model_with_cfg(
+        ConvNeXt, variant, pretrained,
+        default_cfg=default_cfgs[variant],
+        pretrained_filter_fn=checkpoint_filter_fn,
+        feature_cfg=dict(out_indices=(0, 1, 2, 3), flatten_sequential=True),
+        **kwargs)
+    return model
+
+
+@register_model
+def convnext_tiny(pretrained=False, **kwargs):
+    model_args = dict(depths=(3, 3, 9, 3), dims=(96, 192, 384, 768), **kwargs)
+    model = _create_convnext('convnext_tiny', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_tiny_hnf(pretrained=False, **kwargs):
+    model_args = dict(depths=(3, 3, 9, 3), dims=(96, 192, 384, 768), head_norm_first=True, **kwargs)
+    model = _create_convnext('convnext_tiny_hnf', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_small(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[96, 192, 384, 768], **kwargs)
+    model = _create_convnext('convnext_small', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_base(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[128, 256, 512, 1024], **kwargs)
+    model = _create_convnext('convnext_base', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_large(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[192, 384, 768, 1536], **kwargs)
+    model = _create_convnext('convnext_large', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_base_in22k(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[128, 256, 512, 1024], **kwargs)
+    model = _create_convnext('convnext_base_in22k', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_large_in22k(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[192, 384, 768, 1536], **kwargs)
+    model = _create_convnext('convnext_large_in22k', pretrained=pretrained, **model_args)
+    return model
+
+
+@register_model
+def convnext_xlarge_in22k(pretrained=False, **kwargs):
+    model_args = dict(depths=[3, 3, 27, 3], dims=[256, 512, 1024, 2048], conv_mlp=False, **kwargs)
+    model = _create_convnext('convnext_xlarge_in22k', pretrained=pretrained, **model_args)
+    return model
+
+
+