search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.75k stars 7.96k forks source link

Yolov3 weight optimization #4440

Open payal211 opened 4 years ago

payal211 commented 4 years ago

Hi @AlexeyAB

I have trained yolov3 model using 36 classes. Without reference darknet53.conv.74 Model gives best accuracy for both detection and recognition. But the model size is too big. Approx. 250MB. How do I reduce or optimize this model size by making change in yolov3.cfg

It would be a great help. Thanks in advance.

LukeAI commented 4 years ago

use the tiny-yolo model - 25MB

Or if you want something half way, you could try this "slim yolo" https://github.com/PengyiZhang/SlimYOLOv3/tree/master/cfg/3iter-pruned (don't really need to do the whole pruning thing, just use the config.

payal211 commented 4 years ago

Hi @LukeAI
Thanks for suggestions. Tiny yolo is good for detection but in my case it's not good for classification. I will try 'slim yolo' and update you.