Training mtcnn in KITTI for vehicle detect

[hkdqliu]

Hi, @Seanlinx I have trained mtcnn in KITTI for vehicle detect, I only use the samples contain some cars(only for car detect). I founed that the detction result is so bad. When I use the trained P_net to produce the R_net‘s trained samples, the pos: part: neg = 0.8 : 10: 20, the positive samples's percentage is not balanced, what's your advice to change it?

[Xuan Lin]

@hkdqliu hi, the ratio of pos:part:neg I used in training mtcnn Rnet is about 1:1:2.
But I don't think the network structure of mtcnn is suitable for car detection for the aspect ratio of car bboxes varies.

[hkdqliu]

@Seanlinx I changed the structure of mtcnn to 1:2(height : width). the positive samples produced by Pnet is much less than neg/part ones

[ygl365166495]

Maybe you need remove 'convert_to_square' function and change the 'cellsize' of the 'core/MtcnnDetector.py'.