HuggingFace镜像/tf_efficientnet_b7.ns_jft_in1k
模型介绍文件和版本分析
下载使用量0

tf_efficientnet_b7.ns_jft_in1k 模型卡片

一个 EfficientNet 图像分类模型。由论文作者使用 TensorFlow，通过 Noisy Student 半监督学习在 ImageNet-1k 和未标记的 JFT-300m 上进行训练，后由 Ross Wightman 移植到 PyTorch。

模型详情

  • 模型类型： 图像分类 / 特征骨干网络
  • 模型统计：
    • 参数（M）：66.3
    • GMACs：38.3
    • 激活值（M）：289.9
    • 图像尺寸：600 x 600
  • 相关论文：
    • EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks（EfficientNet：重新思考卷积神经网络的模型缩放）：https://arxiv.org/abs/1905.11946
    • Self-training with Noisy Student improves ImageNet classification（使用 Noisy Student 自训练提升 ImageNet 分类效果）：https://arxiv.org/abs/1911.04252
  • 数据集： ImageNet-1k
  • 原始地址： https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet

模型用途

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('tf_efficientnet_b7.ns_jft_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)

开放思维

import torch 
import timm
import argparse
from PIL import Image
from openmind import is_torch_npu_available

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to model",
        default='../',
    )
    args = parser.parse_args()
    return args

if __name__ == "__main__":
    args = parse_args()
    model_path = args.model_name_or_path
    img_path = model_path + '/img/pexels-ioanamtc-7521429.jpg'
    img = Image.open(img_path)

    if is_torch_npu_available():
        device = "npu:0"
    else:
        device = "cpu"

    model_name = 'tf_efficientnet_b7.ns_jft_in1k'
    checkpoint_path=model_path + '/pytorch_model.bin'
    model = timm.create_model(model_name, pretrained=False, checkpoint_path=checkpoint_path).to(device)
    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).to(device))  # unsqueeze single image into batch of 1
    img.close()

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

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(
    'tf_efficientnet_b7.ns_jft_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, 32, 300, 300])
    #  torch.Size([1, 48, 150, 150])
    #  torch.Size([1, 80, 75, 75])
    #  torch.Size([1, 224, 38, 38])
    #  torch.Size([1, 640, 19, 19])

    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(
    'tf_efficientnet_b7.ns_jft_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, 2560, 19, 19) shaped tensor

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

模型对比

在 timm 的 model results 中探索此模型的数据集和运行时指标。

引用

@inproceedings{tan2019efficientnet,
  title={Efficientnet: Rethinking model scaling for convolutional neural networks},
  author={Tan, Mingxing and Le, Quoc},
  booktitle={International conference on machine learning},
  pages={6105--6114},
  year={2019},
  organization={PMLR}
}
@article{Xie2019SelfTrainingWN,
  title={Self-Training With Noisy Student Improves ImageNet Classification},
  author={Qizhe Xie and Eduard H. Hovy and Minh-Thang Luong and Quoc V. Le},
  journal={2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019},
  pages={10684-10695}
}
@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}}
}