安裝darknet
-
詳細步驟和解釋請參考YOLOv3的官網，這裏給出相應的命令行和需要特別注意的地方
- 從github上下載項目源碼
git clone https://github.com/pjreddie/darknet cd darknet - 對源碼進行編譯(編譯默認運行YOLOv3時是不使用GPU的，我們假設跑YOLOv3時需要使用GPU，所以需要修改編譯選項)
- 只需要將當前文件夾下（darknet/）的Makefile文件中第一行的GPU=0改爲GPU=1，改完後的Makefile：
GPU=1 CUDNN=0 OPENCV=0 OPENMP=0 DEBUG=0 ... ... ... -
編譯
make
- 只需要將當前文件夾下（darknet/）的Makefile文件中第一行的GPU=0改爲GPU=1，改完後的Makefile：
- 將權重文件（yolov3.weights）下載到當前目錄下
wget https://pjreddie.com/media/files/yolov3.weights - 測試
#命令行中的‘-i 0’代表使用第0號顯卡，假如不想使用GPU可以用‘-nogpu’替代‘-i 0’ ./darknet -i 0 detector test cfg/coco.data cfg/yolov3.cfg yolov3.weights data/dog.jpg
COCO數據集的下載和配置
- 通過腳本文件（scripts/get_coco_dataset.sh）獲取COCO數據集（數據集下載速度會很慢，建議從Function的數據集下載站下載，然後逐行運行scripts/get_coco_dataset.sh中的每一行命令）
cp scripts/get_coco_dataset.sh data cd data bash get_coco_dataset.sh - 修改COCO的配置文件（cfg/coco.data）
classes= 80 train = <path-to-coco>/trainvalno5k.txt # 注意這裏的<path-to-coco>需要填入COCO數據集的文件夾路徑 valid = <path-to-coco>/5k.txt # 注意這裏的<path-to-coco>需要填入COCO數據集的文件夾路徑 names = data/coco.names backup = backup eval = coco #注意這行是爲了生成json格式的檢測結果，官網裏並沒有說，如果去掉這一行生成的文件是classes(80)類個txt文件 -
修改YOLOv3模型的配置文件（cfg/yolov3.cfg）
[net] # Testing # batch=1 # subdivisions=1 # Training batch=64 #注意這裏的batch有可能使得顯存爆掉,這時需要修改,GeForce RTX 2080Ti需要修改爲8 subdivisions=8 ....
對COCO驗證集進行檢測
-
在darknet目錄下運行下面的命令行
./darknet -i 0 detector valid cfg/coco.data cfg/yolov3.cfg yolov3.weights # '-i 0'代表使用0號GPU運行完上述命令後會生成results/coco_results.json文件，該文件保存了檢測結果
計算mAP
方法源於：https://blog.csdn.net/xidaoliang/article/details/88397280
- 安裝pycocotools
pip install pycocotools -
在darknet目錄下編寫一個python腳本(compute_coco_mAP.py)用於計算mAP（注意根據自己的實際情況修改兩個文件路徑）
#-*- coding:utf-8 -*- import matplotlib.pyplot as plt from pycocotools.coco import COCO from pycocotools.cocoeval import COCOeval import numpy as np import skimage.io as io import pylab,json pylab.rcParams['figure.figsize'] = (10.0, 8.0) def get_img_id(file_name): ls = [] myset = [] annos = json.load(open(file_name, 'r')) for anno in annos: ls.append(anno['image_id']) myset = {}.fromkeys(ls).keys() return myset if __name__ == '__main__': annType = ['segm', 'bbox', 'keypoints']#set iouType to 'segm', 'bbox' or 'keypoints' annType = annType[1] # specify type here cocoGt_file = 'data/coco/annotations/instances_val2014.json' #需要根據自己的實際情況配置該路徑 cocoGt = COCO(cocoGt_file)#取得標註集中coco json對象 cocoDt_file = 'results/coco_results.json' #需要根據自己的實際情況配置該路徑 imgIds = get_img_id(cocoDt_file) print(len(imgIds)) cocoDt = cocoGt.loadRes(cocoDt_file)#取得結果集中image json對象 imgIds = sorted(imgIds)#按順序排列coco標註集image_id imgIds = imgIds[0:5000]#標註集中的image數據 cocoEval = COCOeval(cocoGt, cocoDt, annType) cocoEval.params.imgIds = imgIds#參數設置 cocoEval.evaluate()#評價 cocoEval.accumulate()#積累 cocoEval.summarize()#總結 -
運行compute_coco_mAP.py腳本
python compute_coco_mAP.py -
有可能在運行compute_coco_mAP.py腳本的過程中會報錯，可能的解決方法：
-
升級scikit-image
pip install -U scikit-image - 修改Numpy版本爲1.16
pip install numpy==1.16
-
結果展示
- img_size爲608*608（img_size的設置可以通過修改cfg/yolov3.cfg文件中的width和height來實現）時：
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.334 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.585 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.345 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.194 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.365 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.439 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.291 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.446 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.470 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.304 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.502 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.593 - img_size爲416*416時：
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.314 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.559 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.318 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.142 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.341 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.464 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.278 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.419 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.442 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.239 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.482 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.611 - img_size爲320*320時：
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.286 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.521 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.284 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.103 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.315 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.449 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.260 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.389 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.408 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.189 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.454 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.601
關於結果的解釋請參考：https://blog.csdn.net/u014734886/article/details/78831884