Using r-frcnn

[XWalways]

I want to use the r-frcnn(rotation) to train my own datasets, What should my data format be ? (cx,cy,w,h,angle) or (x0,y0,x1,y1,x2,y2,x3,y3)? But if it is the former, What is the calculation method and representation of the angle? radian or not?

[makefile]

The format is (cx, cy, w, h, θ), θ is anti-clockwise radian.

[XWalways]

for HRSC2016,in the annotation (cx,cy,w,h,angle),w is always larger than h,and the angle is the the angle between widths,Am I right?

[makefile]

Yes, same as human normal sense.

[XWalways]

Thanks a lot,good luck for you

[XWalways]

Excuse me ,I have another question,while training on my own dataset, Check failed: error == cudaSuccess (9 vs. 0) invalid configuration argument occurred ，I've browsed your blog at https://www.cnblogs.com/makefile/p/caffe-errors.html,but it cannot solve the error, I have checked my dataset ,class_num = 5 (0 means bg;1,2,3,4,represent the other four categories ) ,How to slove it?

[makefile]

I'm not sure either, there may be many aspect causing this error. Check your dataset's annotation, and try adjust some of the hyperparams.

[XWalways]

ok,i will check my data again.i have changed the image size,batch size,gpu id,should i change the anchor settings?

[makefile]

You can try it, other thing like iou thresh.

[XWalways]

Sorry to bother you again,the error Check failed: error == cudaSuccess (9 vs. 0) invalid configuration argument occurred has gone ,But the error Check failed: error == cudaSuccess (77 vs. 0) an illegal memory access was encountered appeard,I checked a lot of information, but didn't solve it.