dev(narugo): fix bug about register (78339643) · Commits · git-mirror / Imgutils

requirements-zoo.txt

+1 −1

Original line number	Diff line number	Diff line
		@@ -6,10 +6,10 @@ tqdm
		onnx
		onnxoptimizer
		onnxsim
		onnxruntime
		click
		di-toolkit
		tensorboard
		einops
		thop
		accelerate
		timm
		No newline at end of file

zoo/monochrome/metaformer.py

+8 −4

Original line number	Diff line number	Diff line
		import torch
		from torch import nn
		from timm.models import create_model
		import zoo.monochrome.metaformer_timm # register models

		from .metaformer_timm import __file__ as _bullshit

		_ = _bullshit


		class CAFormerBuilder:
		__model_name__ = 'caformer'
		@@ -28,6 +31,7 @@ class CAFormerBuilder:
		model = create_model(**self.create_model_args)
		return model


		if __name__ == '__main__':
		from thop import profile

zoo/monochrome/metaformer_timm.py

+125 −116

Original line number	Diff line number	Diff line
		@@ -18,12 +18,13 @@ ConvFormer and CAFormer.
		Some implementations are modified from timm (https://github.com/rwightman/pytorch-image-models).
		"""
		from functools import partial

		import torch
		import torch.nn as nn
		from timm.models.layers import trunc_normal_, DropPath
		from timm.models.registry import register_model
		from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
		from timm.models.layers import trunc_normal_, DropPath
		from timm.models.layers.helpers import to_2tuple
		from timm.models.registry import register_model


		def _cfg(url='', **kwargs):
		@@ -48,7 +49,6 @@ default_cfgs = {
		'identityformer_m48': _cfg(
		url='https://huggingface.co/sail/dl/resolve/main/identityformer/identityformer_m48.pth'),


		'randformer_s12': _cfg(
		url='https://huggingface.co/sail/dl/resolve/main/randformer/randformer_s12.pth'),
		'randformer_s24': _cfg(
		@@ -71,8 +71,6 @@ default_cfgs = {
		'poolformerv2_m48': _cfg(
		url='https://huggingface.co/sail/dl/resolve/main/poolformerv2/poolformerv2_m48.pth'),



		'convformer_s18': _cfg(
		url='https://huggingface.co/sail/dl/resolve/main/convformer/convformer_s18.pth'),
		'convformer_s18_384': _cfg(
		@@ -129,7 +127,6 @@ default_cfgs = {
		url='https://huggingface.co/sail/dl/resolve/main/convformer/convformer_b36_in21k.pth',
		num_classes=21841),


		'caformer_s18': _cfg(
		url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18.pth'),
		'caformer_s18_384': _cfg(
		@@ -196,6 +193,7 @@ class Downsampling(nn.Module):
		"""
		Downsampling implemented by a layer of convolution.
		"""

		def __init__(self, in_channels, out_channels,
		kernel_size, stride=1, padding=0,
		pre_norm=None, post_norm=None, pre_permute=False):
		@@ -221,6 +219,7 @@ class Scale(nn.Module):
		"""
		Scale vector by element multiplications.
		"""

		def __init__(self, dim, init_value=1.0, trainable=True):
		super().__init__()
		self.scale = nn.Parameter(init_value * torch.ones(dim), requires_grad=trainable)
		@@ -233,9 +232,11 @@ class SquaredReLU(nn.Module):
		"""
		Squared ReLU: https://arxiv.org/abs/2109.08668
		"""

		def __init__(self, inplace=False):
		super().__init__()
		self.relu = nn.ReLU(inplace=inplace)

		def forward(self, x):
		return torch.square(self.relu(x))

		@@ -244,6 +245,7 @@ class StarReLU(nn.Module):
		"""
		StarReLU: s * relu(x) ** 2 + b
		"""

		def __init__(self, scale_value=1.0, bias_value=0.0,
		scale_learnable=True, bias_learnable=True,
		mode=None, inplace=False):
		@@ -254,6 +256,7 @@ class StarReLU(nn.Module):
		requires_grad=scale_learnable)
		self.bias = nn.Parameter(bias_value * torch.ones(1),
		requires_grad=bias_learnable)

		def forward(self, x):
		return self.scale * self.relu(x) ** 2 + self.bias

		@@ -263,6 +266,7 @@ class Attention(nn.Module):
		Vanilla self-attention from Transformer: https://arxiv.org/abs/1706.03762.
		Modified from timm.
		"""

		def __init__(self, dim, head_dim=32, num_heads=None, qkv_bias=False,
		attn_drop=0., proj_drop=0., proj_bias=False, **kwargs):
		super().__init__()
		@@ -281,7 +285,6 @@ class Attention(nn.Module):
		self.proj = nn.Linear(self.attention_dim, dim, bias=proj_bias)
		self.proj_drop = nn.Dropout(proj_drop)


		def forward(self, x):
		B, H, W, C = x.shape
		N = H * W
		@@ -304,6 +307,7 @@ class RandomMixing(nn.Module):
		self.random_matrix = nn.parameter.Parameter(
		data=torch.softmax(torch.rand(num_tokens, num_tokens), dim=-1),
		requires_grad=False)

		def forward(self, x):
		B, H, W, C = x.shape
		x = x.reshape(B, H * W, C)
		@@ -345,6 +349,7 @@ class LayerNormGeneral(nn.Module):
		IdentityFormer, RandFormer and PoolFormerV2 utilize Modified LayerNorm without bias (bias=False);
		ConvFormer and CAFormer utilizes LayerNorm without bias (bias=False).
		"""

		def __init__(self, affine_shape=None, normalized_dim=(-1,), scale=True,
		bias=True, eps=1e-5):
		super().__init__()
		@@ -370,6 +375,7 @@ class SepConv(nn.Module):
		r"""
		Inverted separable convolution from MobileNetV2: https://arxiv.org/abs/1801.04381.
		"""

		def __init__(self, dim, expansion_ratio=2,
		act1_layer=StarReLU, act2_layer=nn.Identity,
		bias=False, kernel_size=7, padding=3,
		@@ -400,6 +406,7 @@ class Pooling(nn.Module):
		Implementation of pooling for PoolFormer: https://arxiv.org/abs/2111.11418
		Modfiled for [B, H, W, C] input
		"""

		def __init__(self, pool_size=3, **kwargs):
		super().__init__()
		self.pool = nn.AvgPool2d(
		@@ -416,6 +423,7 @@ class Mlp(nn.Module):
		""" MLP as used in MetaFormer models, eg Transformer, MLP-Mixer, PoolFormer, MetaFormer baslines and related networks.
		Mostly copied from timm.
		"""

		def __init__(self, dim, mlp_ratio=4, out_features=None, act_layer=StarReLU, drop=0., bias=False, **kwargs):
		super().__init__()
		in_features = dim
		@@ -441,6 +449,7 @@ class Mlp(nn.Module):
		class MlpHead(nn.Module):
		""" MLP classification head
		"""

		def __init__(self, dim, num_classes=1000, mlp_ratio=4, act_layer=SquaredReLU,
		norm_layer=nn.LayerNorm, head_dropout=0., bias=True):
		super().__init__()
		@@ -451,7 +460,6 @@ class MlpHead(nn.Module):
		self.fc2 = nn.Linear(hidden_features, num_classes, bias=bias)
		self.head_dropout = nn.Dropout(head_dropout)


		def forward(self, x):
		x = self.fc1(x)
		x = self.act(x)
		@@ -465,13 +473,13 @@ class MetaFormerBlock(nn.Module):
		"""
		Implementation of one MetaFormer block.
		"""

		def __init__(self, dim,
		token_mixer=nn.Identity, mlp=Mlp,
		norm_layer=nn.LayerNorm,
		drop=0., drop_path=0.,
		layer_scale_init_value=None, res_scale_init_value=None
		):

		super().__init__()

		self.norm1 = norm_layer(dim)
		@@ -545,6 +553,7 @@ class MetaFormer(nn.Module):
		output_norm: norm before classifier head. Default: partial(nn.LayerNorm, eps=1e-6).
		head_fn: classification head. Default: nn.Linear.
		"""

		def __init__(self, in_chans=3, num_classes=1000,
		depths=[2, 2, 6, 2],
		dims=[64, 128, 320, 512],
		@@ -641,7 +650,6 @@ class MetaFormer(nn.Module):
		return x



		@register_model
		def identityformer_s12(pretrained=False, **kwargs):
		model = MetaFormer(
		@@ -822,7 +830,6 @@ def randformer_m48(pretrained=False, **kwargs):
		return model



		@register_model
		def poolformerv2_s12(pretrained=False, **kwargs):
		model = MetaFormer(
		@@ -1542,6 +1549,7 @@ def caformer_m364_in21k(pretrained=False, **kwargs):
		model.load_state_dict(state_dict, strict=False)
		return model


		@register_model
		def caformer_d30(pretrained=False, **kwargs):
		model = MetaFormer(
		@@ -1553,6 +1561,7 @@ def caformer_d30(pretrained=False, **kwargs):
		model.default_cfg = default_cfgs['caformer_d30']
		return model


		@register_model
		def caformer_b36(pretrained=False, **kwargs):
		model = MetaFormer(

zoo/monochrome/train_.py

+1 −1

Original line number	Diff line number	Diff line
		@@ -18,10 +18,10 @@ from .dataset import MonochromeDataset, Monochrome2DDataset, random_split_datase
		from .levit1d import LeSigTransformer
		from .levit2d import LeViT
		from .loss import FocalLoss
		from .metaformer import CAFormerBuilder
		from .resnet1d import ResNet18, ResNet34, ResNet50, ResNet101, ResNet152
		from .resnet2d import ResNet182D, ResNet342D, ResNet502D, ResNet1012D, ResNet1522D
		from .transformer import SigTransformer
		from .metaformer import CAFormerBuilder
		from ..base import _TRAIN_DIR as _GLOBAL_TRAIN_DIR

		_TRAIN_DIR = os.path.join(_GLOBAL_TRAIN_DIR, 'monochrome')