rizavelioglu/fashionfail

---
license: other
license_name: server-side-public-license
license_link: https://www.mongodb.com/licensing/server-side-public-license
task_categories:
- object-detection
- image-segmentation
tags:
- fashion
- e-commerce
- apparel
size_categories:
- 1K<n<10K
---

# FashionFail Dataset

The FashionFail dataset, proposed in the paper ["FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation"](https://arxiv.org/abs/2404.08582),
(also check the [project page](https://rizavelioglu.github.io/fashionfail/))
comprises 2,495 high-resolution images (2400x2400 pixels) of products found on e-commerce websites. The dataset is divided into training, validation, and test sets, consisting of 1,344, 150, and 1,001 images, respectively.

> Note: The annotations are **automatically** generated by foundation models. However, a human annotator reviewed each sample to ensure the accuracy of the annotations.

### Download Dataset
To address concerns regarding data regulations, we share only the URLs of the images, rather than sharing the image files directly.
However, we provide a simple script to facilitate dataset construction.
The script initially retrieves annotation files from HuggingFace Datasets, then proceeds to download images using the URLs provided in those annotation files.

First, install the repository with:
```
git clone https://github.com/rizavelioglu/fashionfail.git
cd fashionfail
pip install -e .
```

Then, execute the following script:
```
python fashionfail/data/make_dataset.py
```
which constructs the dataset inside `"~/.cache/fashionfail/"`.
An optional argument `--save_dir` can be set to construct the dataset in the preferred directory.

### Annotation format

We follow the annotation format of the [COCO dataset](https://cocodataset.org/#format-data). The annotations are stored in the [JSON format](http://www.json.org/) and are organized as follows:

```
{
"info" : info, # dict: keys are shown below
"licenses" : [license], # List[dict]: keys are shown below
"categories" : [category], # List[dict]: keys are shown below
"images" : [image], # List[dict]: keys are shown below
"annotations" : [annotation], # List[dict]: keys are shown below
}

info{
"year" : int,
"version" : str,
"description" : str,
"contributor" : str,
"url" : str,
"date_created" : datetime,
}

license{
"id" : int,
"name" : str,
"url" : str,
}

category{
"id" : int,
"name" : str,
"supercategory" : str,
}

image{
"id" : int,
"file_name" : str,
"height" : int,
"width" : int,
"license" : int,
"original_url" : str,
}

annotation{
"id" : int,
"image_id" : int,
"category_id" : int,
"area" : int,
"iscrowd" : int, # always 0 as instances represent a single object
"bbox" : list[float], # [x,y,width,height]
"segmentation" : str, # compressed RLE: {"size", (height, widht), "counts": str}
}
```

### License
TL;DR: Not available for commercial use, unless the FULL source code is shared! \
This project is intended solely for academic research. No commercial benefits are derived from it.
All images and brands are the property of their respective owners: © adidas 2023.
Annotations are licensed under [Server Side Public License (SSPL)](https://www.mongodb.com/legal/licensing/server-side-public-license)

### Citation
```
@inproceedings{velioglu2024fashionfail,
author = {Velioglu, Riza and Chan, Robin and Hammer, Barbara},
title = {FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation},
journal = {IJCNN},
eprint = {2404.08582},
year = {2024},
}
```

原始信息汇总

FashionFail 数据集

概述

FashionFail 数据集是由论文 "FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation" 提出的，包含 2,495 张高分辨率（2400x2400 像素）的电子商务网站产品图像。数据集分为训练集、验证集和测试集，分别包含 1,344、150 和 1,001 张图像。

数据集下载

数据集仅提供图像的 URL，而非直接共享图像文件。提供了一个脚本用于数据集构建，该脚本首先从 HuggingFace Datasets 获取注释文件，然后使用这些注释文件中的 URL 下载图像。

安装与使用

1. 克隆仓库并安装： bash git clone https://github.com/rizavelioglu/fashionfail.git cd fashionfail pip install -e .

2. 运行脚本构建数据集： bash python fashionfail/data/make_dataset.py

   数据集将构建在 "~/.cache/fashionfail/" 目录下，可选参数 --save_dir 可设置为指定目录。

注释格式

注释遵循 COCO 数据集 的格式，存储为 JSON 格式，结构如下： json { "info" : info, # dict: keys are shown below "licenses" : [license], # List[dict]: keys are shown below "categories" : [category], # List[dict]: keys are shown below "images" : [image], # List[dict]: keys are shown below "annotations" : [annotation], # List[dict]: keys are shown below }

info{ "year" : int, "version" : str, "description" : str, "contributor" : str, "url" : str, "date_created" : datetime, }

license{ "id" : int, "name" : str, "url" : str, }

category{ "id" : int, "name" : str, "supercategory" : str, }

image{ "id" : int, "file_name" : str, "height" : int, "width" : int, "license" : int, "original_url" : str, }

annotation{ "id" : int, "image_id" : int, "category_id" : int, "area" : int, "iscrowd" : int, # always 0 as instances represent a single object "bbox" : list[float], # [x,y,width,height] "segmentation" : str, # compressed RLE: {"size", (height, widht), "counts": str} }

许可证

该项目仅用于学术研究，不提供商业使用。图像和品牌归各自所有者所有。注释文件根据 Server Side Public License (SSPL) 授权。

引用

bibtex @inproceedings{velioglu2024fashionfail, author = {Velioglu, Riza and Chan, Robin and Hammer, Barbara}, title = {FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation}, journal = {IJCNN}, eprint = {2404.08582}, year = {2024}, }