• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    公众号

junsukchoe/ADL: Attention-based Dropout Layer for Weakly Supervised Object Local ...

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

junsukchoe/ADL

开源软件地址(OpenSource Url):

https://github.com/junsukchoe/ADL

开源编程语言(OpenSource Language):

Python 99.9%

开源软件介绍(OpenSource Introduction):

Attention-based Dropout Layer for Weakly Supervised Object Localization

Attention-based Dropout Layer for Weakly Supervised Object Localization
Junsuk Choe, and Hyunjung Shim
School of Integrated Technology, Yonsei University

Weakly Supervised Object Localization (WSOL) techniques learn the object location only using image-level labels, without location annotations. A common limitation for these techniques is that they cover only the most discriminative part of the object, not the entire object. To address this problem, we propose an Attention-based Dropout Layer (ADL), which utilizes the self-attention mechanism to process the feature maps of the model. The proposed method is composed of two key components: 1) hiding the most discriminative part from the model for capturing the integral extent of object, and 2) highlighting the informative region for improving the recognition power of the model. Based on extensive experiments, we demonstrate that the proposed method is effective to improve the accuracy of WSOL, achieving a new state-of-the-art localization accuracy in CUB-200-2011 dataset. We also show that the proposed method is much more efficient in terms of both parameter and computation overheads than existing techniques.

RubberDuck

ADL block diagram. The self-attention map is generated by channelwise average pooling of the input feature map. Based on the self-attention map, we produce a drop mask using thresholding and an importance map using a sigmoid activation, respectively. The drop mask and the importance map are selected stochastically at each iteration and applied to the input feature map. Please note that this figure illustrates the case when the importance map is selected.

Getting Started

Tensorpack implementation of Attention-Dropout Layer for Weakly Supservised Object Localization.
PyTorch implementation is available at: link

Our implementation is based on these repositories:

Imagenet Pre-trained models can be downloaded here:

Requirements

  • Python 3.3+
  • Python bindings for OpenCV.
  • Tensorflow (≥ 1.12, < 2)

Prepare datasets

ImageNet

To prepare ImageNet data, download ImageNet "train" and "val" splits from here and put the downloaded file on dataset/ILSVRC2012_img_train.tar and dataset/ILSVRC2012_img_val.tar.

Then, run the following command on root directory to extract the images.

./dataset/prepare_imagenet.sh

apt-get install parallel may be required to install parallel.

The structure of image files looks like

dataset
└── ILSVRC
    └── train
        └── n01440764
            ├── n01440764_10026.JPEG
            ├── n01440764_10027.JPEG
            └── ...
        └── n01443537
        └── ...
    └── val
        ├── ILSVRC2012_val_00000001.JPEG
        ├── ILSVRC2012_val_00000002.JPEG
        └── ...

Corresponding annotation files can be found in here.

CUB

Run the following command to download original CUB dataset and extract the image files on root directory.

./dataset/prepare_cub.sh

The structure of image files looks like

dataset
└── CUB
    └── 001.Black_footed_Albatross
        ├── Black_Footed_Albatross_0001_796111.jpg
        ├── Black_Footed_Albatross_0002_55.jpg
        └── ...
    └── 002.Laysan_Albatross
    └── ...

Corresponding annotation files can be found in here.

Training script

First download pretrained models from here. Currently, we provide ResNet50-SE and VGG-16 networks.
Then, run the following command on root directory.

./run_train.sh



鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap