about gaussian yolov3?

[louyanyang]

@eric612 hi,, recently, gaussian yolov3 has a considerable gain, do you have a plan of caffe version?

[eric612]

The fast way is wait AlexeyAB implement and copy his code . 😄 https://github.com/AlexeyAB/darknet/issues/4147

[eric612]

Update , You can reference this project https://github.com/jwchoi384/Gaussian_YOLOv3

[hitzht]

when running Gaussian yolov3,i met a bug in validation as follows : (the bug maybe occurred in the place of solver.cpp:463 ) _*** Aborted at 1574391369 (unix time) try "date -d @1574391369" if you are using GNU date *** PC: @ 0x7f04b0c1a57e (unknown) *** SIGSEGV (@0x7effea7fd4b8) received by PID 19535 (TID 0x7f04b2e38a80) from PID 18446744073348830392; stack trace: *** @ 0x7f04b0b9df20 (unknown) @ 0x7f04b0c1a57e (unknown) @ 0x7f046c17e95f (unknown) @ 0x7f046c23b02e (unknown) @ 0x7f046c23b374 (unknown) @ 0x7f046c23c777 (unknown) @ 0x7f046c1498be (unknown) @ 0x7f046c149bd8 (unknown) @ 0x7f046c2c58e5 cuMemcpy @ 0x7f04b1529ae2 (unknown) @ 0x7f04b1509946 (unknown) @ 0x7f04b152c798 cudaMemcpy @ 0x7f04b2685b35 caffe::caffe_copy<>() @ 0x7f04b271da3d caffe::Yolov3DetectionOutputLayer<>::Forward_gpu() @ 0x7f04b25f1e12 caffe::Net<>::ForwardFromTo() @ 0x7f04b25f1f07 caffe::Net<>::Forward() @ 0x7f04b260c91f caffe::Solver<>::TestDetection() @ 0x7f04b260f3f2 caffe::Solver<>::TestAll() @ 0x7f04b2614a57 caffe::Solver<>::Step() @ 0x7f04b2614c5d caffe::Solver<>::Solve() @ 0x55ce283875a1 train() @ 0x55ce28383893 main @ 0x7f04b0b80b97 __libc_start_main @ 0x55ce283842da start thanks

[eric612]

@hitzht , You can reference my prototxt , but I think it have some training problem , I always got a nan loss when IOU closed to gt
test.txt train.txt solver.txt

[hitzht]

@hitzht , You can reference my prototxt , but I think it have some training problem , I always got a nan loss when IOU closed to gt test.txt train.txt solver.txt

thanks,i will follow your advice to check my code.

[hitzht]

@eric612 I'm also got a nan loss using adam/RMSProp with multistep/triangular2,have you settled this bug?

[AnmachenGuo]

@eric612 I'm also got a nan loss using adam/RMSProp with multistep/triangular2,have you settled this bug?

i also have this problem

[AnmachenGuo]

@hitzht do you have any information abot this problem?