yolo2_light icon indicating copy to clipboard operation
yolo2_light copied to clipboard

Published 20 hours ago • verified publisher icon AlexeyAB

Quantization of the 1st layer

Open daniel89710 opened this issue 5 years ago • 2 comments

Why "disable Quantized for 1st and last layers" in "parse_convolutional"? Quantization for 1st layer is seem to be important for acceleration.

daniel89710 avatar Oct 15 '18 09:10 daniel89710

The 1st and the last layers greatly affect accuracy. So it is done to loss not more than ~1% of mAP by using INT8.

AlexeyAB avatar Oct 15 '18 10:10 AlexeyAB

Special thanks.

daniel89710 avatar Oct 16 '18 10:10 daniel89710