OW-YOLO
OW-YOLO copied to clipboard
Published
20 hours ago •
buxihuo
buxihuo
Detect known and unknown objects in the open world（具有区分已知与未知能力的全新检测器））
简介（Introduction）
赋予检测器初步认知能力，区分已知与未知物体，已实现的检测器有:yolov5
Give the detector a preliminary cognitive ability to distinguish between known and unknown objects.
'''
此为新版本，正在合并用于测试和推理的参数，即将完成，旧版见OW-yolov5-6.0分支。
'''
使用方法（Usage）
Usage 1
Usage 2
Usage 3
快速开始
1. 安装
2. 推理
3. 测试
4. 训练
性能（performance）
1. 精度曲线（precision curve）
说明
模型在coco(80类)数据集训练，在object365(365类)进行测试,将coco上对应的类作为已知类，其他的作为未知类。其中灰色是已知类别的精度曲线，
红色为未知类别精度曲线，蓝色代表所有 类别的平均值。
2. 召回率曲线（recall curve）
3. map
1) coco数据集性能对比
| Model | Param | Flops | mAPval 0.5:0.95 |
mAPval 0.5 |
|---|---|---|---|---|
| Detr | 34 | 78 | 39. 4 | - |
| OW-detr | - | - | 33.1(-6.3) | - |
| yolov5s | 38.46 | 56.8 | ||
| OW-yolov5s(only known) | 38.46 | 57.16 | ||
| OW-yolov5s (unknown background confidence = 0.001) |
37.36 | 54.82 | ||
| OW-yolov5s (unknown background confidence = 0.01) |
38.29 | 56.80 | ||
| OW-yolov5s (unknown background confidence = 0.05) |
38.42 | 57.09 | ||
| OW-yolov5s (unknown background confidence = 0.1) |
38.44 | 57.14 | ||
| OW-yolov5s (unknown background confidence = 0.25) |
38.45 | 57.16 | ||
| OW-yolov5s (unknown background confidence = 0.5) |
38.46 | 57.17 |
2) object365 数据集
| Model | unknown recall | unknown ap50 | all mAPval 0.5:0.95 |
all mAP 0.5 |
|---|---|---|---|---|
| yolov5s | - | - | ||
| OW-yolov5s(only known) | - | - | ||
| OW-yolov5s (unknown background confidence = 0.001) |
0.61 | 0.072 | 22.1 | 31.7 |
| OW-yolov5s (unknown background confidence = 0.01) |
||||
| OW-yolov5s (unknown background confidence = 0.05) |
||||
| OW-yolov5s (unknown background confidence = 0.1) |
0.23 | 12.3 | 22.2 | 32.3 |
| OW-yolov5s (unknown background confidence = 0.25) |
0.15 | 13.5 | 22.2 | 32.3 |
| OW-yolov5s (unknown background confidence = 0.5) |
0.08 | 13.5 | 22.2 | 32.3 |
buxihuo