Tossy0423/yolov4-for-darknet_ros

darknet_ros dies / crash when deleting labels in yolo yaml

SteveJos opened this issue · 1 comments

SteveJos commented

Hey @Tossy0423 ,

great job updating darket_ros to yolov4!
I would like to detect only one class in the image but unfortunately if you remove at least one label in yolov4 (3 as well).yaml darknet crashes exactly in the moment of the first detection.

Here the gdb output:

`[New Thread 0x7ffddffff700 (LWP 28236)]
[Thread 0x7ffde4b9c700 (LWP 28235) exited]
[ INFO] [1603404568.634471341]:

[ INFO] [1603404568.634548849]:
FPS:2101.4

[ INFO] [1603404568.634566248]: Objects:

Thread 42 "darknet_ros" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7ffddffff700 (LWP 28236)]
get_actual_detections (dets=dets@entry=0x7ffdc1b582a0, dets_num=, thresh=thresh@entry=0.300000012, selected_detections_num=selected_detections_num@entry=0x7ffddfffc4ac,
names=names@entry=0x555555e1aab0) at src/darknet_ros/darknet/src/image.c:300
300 int show = strncmp(names[j], "dont_show", 9);
(gdb) [darknet_ros-1] killing on exit
`

Please check this issue....

Thanks

Tossy0423 commented

Hi! @SteveJos

Thank you for use yolov4-for-darknet_ros!

It may be difficult to change the number of objects recognized by simply changing the labels. This is because the model we use in our repository is already trained. If it recognizes only one label, as it did in this case, it may need to be relearned.
It seems there are many ways to relearn yolov4.