File size: 3,180 Bytes
a1c6f00 fb27d38 eef883d fb27d38 eef883d fb27d38 8d7caf4 eef883d 8d7caf4 eef883d 8d7caf4 eef883d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 |
# COHI-O365: A Benchmark Dataset for Fisheye Object Detection
## Dataset Summary
This work introduces COHI-O365, a benchmark dataset for object detection in hemispherical/fisheye images, designed for field-of-view invariant applications. It complements a synthetic training dataset, RMFV365, created by applying fisheye transformations to the Objects365 dataset. COHI-O365 contains 1,000 real fisheye images with 74 classes and an average of 20,798 object instances per image. These images were captured using an ELP-USB8MP02G-L180 hemispherical camera (2448x3264 pixels) and manually annotated with axis-aligned bounding boxes. The RMFV365 dataset, used for model training, comprises 5.1 million fisheye images generated from Objects365. YOLOv7 models were trained on Objects365, RMFV365, and a variant (RMFV365-v1), and evaluated on COHI-O365.
## Dataset Contents
The dataset includes:
* **COHI-O365:** A benchmark testing dataset with 1,000 real fisheye images of 74 classes.
* **RMFV365:** A large-scale synthetic fisheye dataset derived from Objects365, containing 5.1 million images.
A visualization of sample images from both datasets is provided in the GitHub repository. A table detailing the number of bounding boxes per class in COHI-O365 is planned for future inclusion.
## Benchmarks
YOLOv7 models were trained on different datasets and evaluated on COHI-O365. The results are summarized below:
| S/N | Model | Objects365 mAP50 | Objects365 mAP50:95 | RMFV365-v1 mAP50 | RMFV365-v1 mAP50:95 | RMFV365 mAP50 | RMFV365 mAP50:95 | COHI-365 mAP50 | COHI-365 mAP50:95 |
|-----|-------------|--------------------|----------------------|--------------------|---------------------|-----------------|--------------------|-----------------|--------------------|
| 1 | FPN | **35.5** | 22.5 | N/A | N/A | N/A | N/A | N/A | N/A |
| 2 | RetinaNet | 27.3 | 18.7 | N/A | N/A | N/A | N/A | N/A | N/A |
| 3 | YOLOv5m | 27.3 | 18.8 | 22.6 | 14.1 | 18.7 | 10.1 | 40.4 | 28.0 |
| 4 | YOLOv7-0 | 34.97 | **24.57** | 29.1 | 18.3 | 24.2 | 13.0 | 47.5 | 33.5 |
| 5 | YOLOv7-T1 | 34.3 | 24.0 | 32.7 | 22.7 | 32.0 | 22.0 | 49.1 | 34.6 |
| 6 | YOLOv7-T2 | 34 | 23.1 | **32.9** | **23** | **33** | **22.8** | **49.9** | **34.9** |
**Table:** Object recognition results on Objects365, RMFV365-v1, RMFV365, and COHI-365 testing sets. Bold values represent the best performance within each column.
## GitHub Repository
[https://github.com/IS2AI/COHI-O365](https://github.com/IS2AI/COHI-O365)
|