基于darknet的目标检测开发说明书

1. 测试darknet

1) 利用python版本的darknet清楚YOLO V3的输出.

其中要修改一下加载meta数据集, 即*.data中的*.names文件路径. 在测试过程中, 我们将用到meta.classes, meta.names. 这些信息都在*.data中定义了. 经过测试, 对于一张头像而言, YOLO V3的输出格式为:

[('car', 0.9995980858802795, (944.3706665039062, 470.93316650390625, 332.6497802734375, 180.39739990234375)), ('car', 0.9973964691162109, (60.67174530029297, 443.29925537109375, 112.2964553
8330078, 69.57225036621094)), ('traffic light', 0.9912506341934204, (845.8016357421875, 249.35389709472656, 14.662422180175781, 32.508026123046875)), ('car', 0.9833781123161316, (154.60037231445312, 427.7644348144531, 124.64006042480469, 62.710933685302734)), ('traffic light', 0.9458598494529724, (471.6688537597656, 37.09953308105469, 35.19504165649414, 71.88548278808594)), ('traffic light', 0.9146334528923035, (178.1991424560547, 57.963218688964844, 44.318084716796875, 74.83644104003906)), ('traffic light', 0.8367633819580078, (666.54296875, 238.9544677734375, 14.19233226776123, 33.048221588134766)), ('car', 0.8113696575164795, (521.5132446289062, 419.3601989746094, 303.445556640625, 244.7392120361328)), ('traffic light', 0.765159547328949, (1141.6649169921875, 327.1693420410156, 15.563911437988281, 39.063541412353516))]

具体的输出语句为:
(meta.names[i], dets[j].prob[i], (b.x, b.y, b.w, b.h)). 其中, (x, y)是bbox的中点, w是bbox的宽, h是bbox的高.
但是图像显示上会很麻烦, 因为源码是C++写的, 只是提供了Python接口. 显示图像可以采用./darkn