search

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms
GNU General Public License v2.0
718 stars 208 forks source link

reach max boxes #283

Open mochechan opened 2 years ago

mochechan commented 2 years ago

My detecting program uses tkdnn with yolov4 model, and it is good for many pictures. One day, my detecting program crashes at "reach max boxes".

reach max boxes
/home/a/tkDNN/src/Yolo.cpp:171
Aborting...
[Thread 0x7f96d0dc60 (LWP 6878) exited]
[Thread 0x7f5b7fcc60 (LWP 6887) exited]
[Thread 0x7f68c98c60 (LWP 6885) exited]
[Thread 0x7f69499c60 (LWP 6884) exited]
[Thread 0x7f9df8b010 (LWP 6875) exited]
[Thread 0x7f942ddc60 (LWP 7454) exited]
[Thread 0x7f95703c60 (LWP 7453) exited]
[Thread 0x7f5affbc60 (LWP 7457) exited]
[Thread 0x7f68497c60 (LWP 7456) exited]
[Thread 0x7f907c5c60 (LWP 7455) exited]
[Thread 0x7f917c7c60 (LWP 6882) exited]
[Thread 0x7f90fc6c60 (LWP 6883) exited]
[Thread 0x7f91fc8c60 (LWP 6881) exited]
[Inferior 1 (process 6875) exited with code 01]
No stack.

I follow the source code to know the reason being that "MAX_DETECTIONS" should be too small. The strange is that MAX_DETECTIONS is already 8192*2. Is it a good solution to enlarge MAX_DETECTIONS? Is it producing some side-effect to enlarge MAX_DETECTIONS? Any other good advice?

ps. "try catch" cannot ignore this crash.

gewenpulan commented 1 year ago

I also have same problem. In my case, the camera watching same place which object on video almost same, and 99%image predict good,but sometimes the model return manymany rectance like [760, 709, 760, 709] [1000, 720, 1000, 720][-5368709120, -3019898880, -5368709120, -3019898880] and i dont know why