regnetz_d8_evos.ch_in1k模型卡片

一个RegNetZ图像分类模型。由Ross Wightman在timm中使用ImageNet-1k进行训练。

这些RegNetZ B / C / D模型探索了不同的组大小和层配置，并没有遵循任何论文描述。与EfficientNets类似，该架构使用线性（非激活的）块输出和反转的瓶颈（中间块扩展）。

• B16：~1.5GF @ 256x256，组宽度为16。单层干扰。
• C16：~2.5GF @ 256x256，组宽度为16。单层干扰。
• D32：~6GF @ 256x256，组宽度为32。分层3层干扰，无池化。
• D8：~4GF @ 256x256，组宽度为8。分层3层干扰，无池化。
• E8：~10GF @ 256x256，组宽度为8。分层3层干扰，无池化。

该模型使用定制的EvoNorm-S0归一化-激活层，而不是BatchNorm与SiLU激活。

该模型架构是使用timm的灵活的 BYOBNet (Bring-Your-Own-Blocks Network) 实现的。

BYOBNet允许配置：

• 块/阶段布局
• 干扰布局
• 输出步幅（扩张）
• 激活和规范化层
• 通道和空间/自注意层

...还包括timm中许多其他架构的特征，包括：

• 随机深度
• 梯度检查点
• 层次化LR衰减
• 每个阶段的特征提取

模型细节

• 模型类型：图像分类/特征抽取
• 模型统计：
  • 参数（M）：23.5
  • GMACs：4.5
  • 激活数（M）：24.9
  • 图像大小：训练= 256 x 256，测试= 320 x 320
• 论文：
  • 快速准确的模型缩放: https://arxiv.org/abs/2103.06877
  • 进化的归一化-激活层: https://arxiv.org/abs/2004.02967
• 数据集：ImageNet-1k
• 原始论文： https://github.com/huggingface/pytorch-image-models

模型用法

图像分类

from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model('regnetz_d8_evos.ch_in1k', pretrained=True)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # unsqueeze single image into batch of 1

top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)

特征图提取

from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model(
    'regnetz_d8_evos.ch_in1k',
    pretrained=True,
    features_only=True,
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # unsqueeze single image into batch of 1

for o in output:
    # print shape of each feature map in output
    # e.g.:
    #  torch.Size([1, 64, 128, 128])
    #  torch.Size([1, 64, 64, 64])
    #  torch.Size([1, 128, 32, 32])
    #  torch.Size([1, 256, 16, 16])
    #  torch.Size([1, 1792, 8, 8])

    print(o.shape)

图像嵌入

from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model(
    'regnetz_d8_evos.ch_in1k',
    pretrained=True,
    num_classes=0,  # remove classifier nn.Linear
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # output is (batch_size, num_features) shaped tensor

# or equivalently (without needing to set num_classes=0)

output = model.forward_features(transforms(img).unsqueeze(0))
# output is unpooled, a (1, 1792, 8, 8) shaped tensor

output = model.forward_head(output, pre_logits=True)
# output is a (1, num_features) shaped tensor

模型比较

在timm model results 中探索此模型的数据集和运行时指标。

对于下面的比较摘要，以ra_in1k、ra3_in1k、ch_in1k、sw_*和lion_*标记的权重在timm中进行训练。

引用

@misc{rw2019timm,
  author = {Ross Wightman},
  title = {PyTorch Image Models},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  doi = {10.5281/zenodo.4414861},
  howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
}

@InProceedings{Dollar2021,
  title = {Fast and Accurate Model Scaling},
  author = {Piotr Doll{'a}r and Mannat Singh and Ross Girshick},
  booktitle = {CVPR},
  year = {2021}
}

@article{liu2020evolving,
  title={Evolving normalization-activation layers},
  author={Liu, Hanxiao and Brock, Andy and Simonyan, Karen and Le, Quoc},
  journal={Advances in Neural Information Processing Systems},
  volume={33},
  pages={13539--13550},
  year={2020}
}