saswat0 / License-Plate-Recognition

Detection of license plate and recognising the registration number
Other
23 stars 12 forks source link

How many frames can you process in one second #4

Closed MingRongXi closed 3 years ago

MingRongXi commented 3 years ago

Hi, saswat0. Thank you for your work, it's pretty! I want to ask you some problem. My environment is 8g GRID M10-8Q, with only one graphics card. If I use darknet and tensorflow compiled with gpu at the same time, the program will report an error: Princess finished with exit code 137 (interrupted by signal 9: SIGKILL). So my plate-detection is so slow, but vehicle detection and ocr are fast. Have you encountered such a problem, and how many frames can you process in one second?

saswat0 commented 3 years ago

Hi @MingRongXi This signal kill is most probably due to insufficient GPU memory. Try running it in CPU instead (enable openmp in the darknet makefile for multiprocessing). The license plate detection is carried out by wpod-net and it's keras implementation is laggy. Vehicle detection and ocr are YOLO models and hence faster. I was able to achieve a FPS of 4 in my i7, 32G (no GPU) system

MingRongXi commented 3 years ago

Thank you!I try it again.