justinhzy / pedestrian-detector Goto Github PK

Hello, sorry to bother you. I want to train on my own dataset to detect pedestrian and cyclist, how I can do it? Thanks!

I am a new learner in this field.
I want to train your model with cityperson
but i don't know how to construct and preprocess my dataset.
thank you

Thanks for your work.Are you chinese?could I add your wechat or QQ?

@Justinhzy Hi, as you mentioned in the Readme, you get the miss rate 15.4%, I didn't find where you calculate it. Would you please specify in which code you get miss rate?
Thanks~

I wanted to know that during your training and testing, did you use the original input image scale i-e im_scale=1 or 1.3 or the default scale of faster rcnn (600 shorter side ) of faster rcnn ? because as per cityperson paper the input image is resized to 1.3 x . Could you please clarify ?

I test the attached model "faster_rcnn_1_8_3759.pth" but get 18.98 % miss rate based on the original image size and default hyper-parameters. Could you tell me the 15.4% settings?