HOD

A Benchmark Dataset for Harmful Object Detection

• This repository provides Harmful Object Dataset and PyTorch implementations for Detection models (YOLOv5 and Faster-RCNN).
• (News) This work is accepted to the WACV 2024 workshop on Real-World Surveillance: Applications and Challenges, 4th.

Authors

• Eungyeom Ha, Heemook Kim, Dongbin Na

Abstract

Recent multi-media data such as images and videos have been rapidly spread out on various online services such as social network services (SNS). With the explosive growth of online media services, the number of image content that may harm users is also growing exponentially. Thus, most recent online platforms such as Facebook and Instagram have adopted content filtering systems to prevent the prevalence of harmful content and reduce the possible risk of adverse effects on users. Unfortunately, computer vision research on detecting harmful content has not yet attracted attention enough. Users of each platform still manually click the report button to recognize patterns of harmful content they dislike when exposed to harmful content. However, the problem with manual reporting is that users are already exposed to harmful content. To address these issues, our research goal in this work is to develop automatic harmful object detection systems for online services. We present a new benchmark dataset for harmful object detection. Unlike most related studies focusing on a small subset of object categories, our dataset addresses various categories. Specifically, our proposed dataset contains more than 10,000 images across 6 categories that might be harmful, consisting of not only normal cases but also hard cases that are difficult to detect. Moreover, we have conducted extensive experiments to evaluate the effectiveness of our proposed dataset. We have utilized the recently proposed state-of-the-art (SOTA) object detection architectures and demonstrated our proposed dataset can be greatly useful for the real-time harmful object detection task.

Datasets

• The dataset is divided into two distinct groups based on the difficulty of detection: the Normal cases and the Hard cases.

  

• This repository provides (1) normal cases dataset, (2) hard cases dataset per category.

        Dataset/
            alcohol/
                normal/
                hard/
            insulting_gesture/
                normal/
                hard/
            .
            .
            .
                            
            knife/
                normal/
                hard/
    

• The number of images for each case per category.

  	alcohol	insulting gesture	blood	cigarette	gun	knife
  normal cases	533	466	554	550	999	2,366
  hard cases	978	267	994	1,538	566	820

Source Codes

• Datasets and source codes that match the experimental environment.

  	YOLOv5	Faster R-CNN
  dataset	YOLOv5 dataset zip	Faster R-CNN dataset zip
  all code	YOLOv5 code	Faster-RCNN Training code Faster-RCNN Test code: Normal Cases Faster-RCNN Test code: Hard Cases
  normal code	YOLOv5 code	Faster-RCNN Training code Faster-RCNN Test code: Normal Cases Faster-RCNN Test code: Hard Cases

Demonstration

• The example images from the hard case test dataset and the corresponding inference results based on the training dataset.

  

• The detection performance of trained models.

  

Citation

If this work can be useful for your research, please cite our paper:

@misc{ha2023hod,
      title={HOD: A Benchmark Dataset for Harmful Object Detection}, 
      author={Eungyeom Ha and Heemook Kim and Sung Chul Hong and Dongbin Na},
      year={2023},
      eprint={2310.05192},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Notice Regarding Dataset Usage

Dear Researchers and Collaborators,

As part of our commitment to advancing research, we have provided a benchmark dataset associated with our recent publication. While we are excited to share this resource with the community, we want to highlight some important considerations:

• Copyright Compliance: This dataset is shared for research purposes only. Please ensure that your use of the dataset complies with all applicable copyright laws and regulations. Redistribution of the dataset without proper authorization is not permitted.

• Privacy Considerations: We have taken steps to ensure the privacy and confidentiality of any individuals or entities represented in the dataset. Users of this dataset are expected to uphold these standards and not use the data in any way that compromises individual privacy.

• Intended Use: This dataset is intended solely for academic and research purposes. We urge users to apply this data responsibly and ethically, keeping in mind its intended application in the field.

• User Agreement: By accessing and using this dataset, users agree to adhere to these terms and conditions. Misuse of the data or violation of these guidelines may result in restricted access to future resources.

We appreciate your cooperation in using this dataset responsibly. If you have any concerns or questions regarding the dataset, please feel free to contact us through GitHub.

Thank you for supporting responsible research practices.