# COHI-O365: A Benchmark Dataset for Fisheye Object Detection

**Repository:** [https://github.com/IS2AI/COHI-O365](https://github.com/IS2AI/COHI-O365)


This repository introduces COHI-O365, a benchmark dataset for object detection in hemispherical/fisheye images, designed for field-of-view invariant applications.  It also includes the RMFV365 dataset, a large-scale synthetic fisheye dataset used for training.  Pre-trained YOLOv7 models are provided, trained on various combinations of these datasets.


## Dataset Summary

COHI-O365 is a real-world dataset containing 1,000 fisheye images of 74 classes, sampled from the Objects365 dataset.  These images, captured using an ELP-USB8MP02G-L180 hemispherical camera (2448x3264 resolution), feature an average of 20,798 object instances per image and are annotated with axis-aligned bounding boxes.


RMFV365 is a synthetic fisheye dataset created by applying non-linear mapping to the Objects365 dataset.  It comprises 5.1 million images offering a diverse range of perspectives and distortions, useful for training robust object detection models.


<img src="https://github.com/IS2AI/COHI-O365/blob/main/pictures/COHI-365 Sample Images.png" width="750">

*COHI-O365 Sample Images*


<img src="https://github.com/IS2AI/COHI-O365/blob/main/pictures/RMFV365 Sample Images.png" width="750">

*RMFV365 Sample Images*


## Pre-trained Models and Results

Three YOLOv7 models were trained and evaluated:

* **YOLOv7-0:** Trained on Objects365.
* **YOLOv7-T1:** Trained on Objects365 and a variant of RMFV365 (RMFV365-v1), using a lens and camera-independent fisheye transformation (n=4).
* **YOLOv7-T2:** Trained on RMFV365.


<img src="https://github.com/IS2AI/COHI-O365/blob/main/pictures/training_scheme.jpg" width="500">

*Model Training Scheme*


The models' performance, measured by mAP<sub>50</sub>, is summarized below:

<table>
    <thead>
        <tr>
            <th rowspan="3">S/N</th>
            <th rowspan="3">Model</th>
            <th colspan="8">Test Results (%)</th>
        </tr>
        <tr>
            <th colspan="2">Objects365</th>
            <th colspan="2">RMFV365-v1</th>
            <th colspan="2">RMFV365</th>
            <th colspan="2">COHI-365</th>
        </tr>
        <tr>
            <th>mAP50</th>
            <th>mAP50:95</th>
            <th>mAP50</th>
            <th>mAP50:95</th>
            <th>mAP50</th>
            <th>mAP50:95</th>
            <th>mAP50</th>
            <th>mAP50:95</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>1</td>
            <td>FPN</td>
            <td><strong>35.5</strong></td>
            <td>22.5</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
        </tr>
        <tr>
            <td>2</td>
            <td>RetinaNet</td>
            <td>27.3</td>
            <td>18.7</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
            <td>N/A</td>
        </tr>
        <tr>
            <td>3</td>
            <td>YOLOv5m</td>
            <td>27.3</td>
            <td>18.8</td>
            <td>22.6</td>
            <td>14.1</td>
            <td>18.7</td>
            <td>10.1</td>
            <td>40.4</td>
            <td>28.0</td>
        </tr>
        <tr>
            <td>4</td>
            <td>YOLOv7-0</td>
            <td>34.97</td>
            <td><strong>24.57</strong></td>
            <td>29.1</td>
            <td>18.3</td>
            <td>24.2</td>
            <td>13.0</td>
            <td>47.5</td>
            <td>33.5</td>
        </tr>
        <tr>
            <td>5</td>
            <td>YOLOv7-T1</td>
            <td>34.3</td>
            <td>24.0</td>
            <td>32.7</td>
            <td>22.7</td>
            <td>32.0</td>
            <td>22.0</td>
            <td>49.1</td>
            <td>34.6</td>
        </tr>
        <tr>
            <td>6</td>
            <td>YOLOv7-T2</td>
            <td>34</td>
            <td>23.1</td>
            <td><strong>32.9</strong></td>
            <td><strong>23</strong></td>
            <td><strong>33</strong></td>
            <td><strong>22.8</strong></td>
            <td><strong>49.9</strong></td>
            <td><strong>34.9</strong></td>
        </tr>
    </tbody>
</table>

**Table:** Objects Recognition Results on Objects365, RMFV365-v1, RMFV365, and COHI-365 Testing Sets


## Citation

*(Citation information not provided in the source content)*