R-CenterNet icon indicating copy to clipboard operation
R-CenterNet copied to clipboard

Published 20 hours ago • verified publisher icon ZeroE04

Problem of angle prediction

Open Fly-dream12 opened this issue 4 years ago • 3 comments

Thanks for you r code. I have trained this model on my dataset, but when i predict the targets, the angle prediction turns to be zero. Furthermore, the loss did not converge when epoch=100. Please give some suggestions.

Fly-dream12 avatar Sep 26 '20 01:09 Fly-dream12

I'm very sorry to hear from you so late. I guess there's something wrong with your label, and i have pushed the labelGenerator to here: https://github.com/ZeroE04/R-CenterNet/tree/master/labelGenerator

you can have a look, good luck for you!

ZeroE04 avatar Oct 21 '20 13:10 ZeroE04

Sorry, I have made the label as you suggested as retrained it, but the results are still with horizontal results. And the angle prediction is still zero.

Fly-dream12 avatar Oct 31 '20 14:10 Fly-dream12

@Fly-dream12 I also meet the same problem. I counted the angles of all my samples as below(ps:I changed the angle interval to (-180,180)): 20210520154343

Could it be caused by the unbalanced angular frequency of the sample? @ZeroE04

Wang-Qinyu avatar May 20 '21 07:05 Wang-Qinyu