msn199959 / logo-2k-plus-dataset Goto Github PK

View Code? Open in Web Editor NEW

106.0 2.0 24.0 772 KB

Logo-2k-plus-Dataset

Python 99.76% Shell 0.24%

logo-2k-plus-dataset's Introduction

Logo-2K+:A Large-Scale Logo Dataset for Scalable Logo Classiﬁcation

The dataset link has been shifted and maintained by this URL https://github.com/Wangjing1551/Logo-2k-plus-Dataset

Logo-2k+ Dataset

Logo-2k+ Dataset Description

In this work, we construct a large scale logo dataset, Logo-2K+, which covers a diverse range of logo classes from real-world logo images. Our resulting logo dataset contains 167,140 images with 10 root categories and 2,341 categories.
The statistic comparison of 10 root categories from Logo-2K+ is shown as follows.

Root Category	Logos	Images
Food	769	54,507
Clothes	286	20,413
Institution	238	17,103
Accessories	210	14,569
Transportation	203	14,719
Electronic	191	13,972
Necessities	182	13,205
Cosmetic	115	7,929
Leisure	99	7,338
Medical	48	3,385
Total	2,341	167,140

Download links

Baidu Drive link: https://pan.baidu.com/s/1L_9JROsWSQEAznNiy-wTBQ password: 945w

Google Drive link: https://drive.google.com/open?id=1PTA24UTZcsnzXPN1gmV0_lRg3lMHqwp6

DRNA-Net

This is a PyTorch implementation of the AAAI2020 paper "Logo-2K+: A Large-Scale Logo Dataset for Scalable Logo Classification".

Requirements

Python >= 3
PyTorch >= 0.4 Install PyTorch >=0.4 with GPU (code are GPU-only), refer to official website
Install cupy, you can install via pip install cupy-cuda80 or(cupy-cuda90,cupy-cuda91, etc).
Install other dependencies: pip install -r requirements.txt

Datasets

Download the Logo-2K+(https://github.com/msn199959/Logo-2k-plus-Dataset) datasets and put it in the root directory. You can also try other classification datasets.

Training on Logo-2K+ dataset:

Download the training, testing data. Since the program loading the data in drna_master/data by default, you can set the data path as following.

cd drna_master
mkdir data
cd data
ln -s $ dataset path

Then you can set some hyper-parameters in drna_master/config.py. If you want to train the DRNA-Net, just run python train.py. During training, the log file and checkpoint file will be saved in save_dir directory.

Test the model

If you want to test the DRNA-Net, just run python test.py. You need to specify the test_model in config.py to choose the checkpoint model for testing.

Reference

If you are interested in our work and want to cite it, please acknowledge the following paper:

@inproceedings{Wang2020Logo2K,
author={Jing Wang, and Weiqing Min, and Sujuan Hou, and Shengnan Ma, and Yuanjie Zheng, and Haishuai Wang, and Shuqiang Jiang},
booktitle={AAAI Conference on Artificial Intelligence. Accepted},
title={{Logo-2K+:} A Large-Scale Logo Dataset for Scalable Logo Classification},
year={2020}
}

logo-2k-plus-dataset's People

Contributors

Stargazers

Watchers

logo-2k-plus-dataset's Issues

error encountered when running train.py

I get the following error when running train.py when I specify to use resnet:
Traceback (most recent call last):
File "train.py", line 56, in
net = model.attention_net(topN=PROPOSAL_NUM)
File "/home/l/liny/Logo-2k-plus-Dataset/Drna-master/core/model_resnet.py", line 75, in init
self.aug_net = aug_net()
File "/home/l/liny/Logo-2k-plus-Dataset/Drna-master/core/model_resnet.py", line 39, in init
self.features = net.get_features()
AttributeError: 'NoneType' object has no attribute 'get_features'

core.utils

Hi, in your train.py, you are trying to import from core.utils import init_log, progress_bar, may I know where is your utils.py ?

Add a license of the code and dataset

Dear Contributor,

I would like to reuse your code and dataset, but without a license I am not sure if you agree and to what extent.

Could you add a licence to the repository?

Source code

A comprehensive info about software licenses can be found here, this one is often used for this kind of projects. It is enough to copy the text and put in LICENSE file in the main dir of the repository.

Data

A comprehensive info about data licenses can be found here. You may want to use this simple CC BY 4.0. It is enough to edit the readme by adding (next to link to the dataset) this words:

The dataset is licensed under a <your chosen license> <link to the license>. Cite is as <title, your name>.

Thank you in advance!

Image Files and Reference TEXT files are out of sync

Images are in folders having names both in English and Chinese, translation doesn't work for finding the exact folder for the class
Image file names and filenames in images.txt and train_test_lable.txt both are out of order numerically

Questions:

How can we parse the txt file to extract and read images?
Is there a file which supports reading both reference txt files and the image files.

70.7%/72.0% for NTS-Net ResNet50/ResNet152

Thanks to the authors to release the Logo-2K+ dataset.

In case one might be interested, we've trained the NTS-Net on this dataset, achieving ~70.7% with ResNet50 and ~72% with ResNet152. See https://github.com/elv-youliangyu/logo for details.

Edit: updated the latest training results for NTS-Net ResNet50/ResNet152

release code for baselines

Hello,

Could you please release the code for how you trained the baselines (esp. resnet-50)? Training a resnet-50 (pretrained on imagenet) gives only 13% accuracy for me. Did you train one from scratch for what's in the paper?

original resolution images

In your Logo-2K+ dataset, all the images seem to be resized to 256x256x3.
Is it possible to provide the same image set with original resolution images?
Thank you in advance.

is it possible to recognize several logo instances ?

Hi,

Hope you are all well !

Is it possible to detect several logo instances like with yolo with your code (I know not yet released ^^) ?

Thanks for any inputs or insights on that topic.

Cheers,
X