the speed of training about lffd-a-light-and-fast-face-detector-for-edge-devices HOT 4 OPEN

@huanpig pkl format is much faster than rec of mxnet. pkl files are loaded into the memory, avoiding hard disk IO. The speed is determined by factors: 1) jpg decode speed (cv2 with turbo-jpeg); 2) the image size and the input size; 3) the average number of objects in input images (for loops).
So the CPU performance matters.

from lffd-a-light-and-fast-face-detector-for-edge-devices.

in article,you say:We train
1,500,000 iterations and reduce the learning rate by multiplying 0.1 at iteration 600,000, 1,000,000, 1,200,000 and
1,400,000. The training time is about 5 days with two
NVIDIA GTX1080TI, in my thought,it's very slow, is it uner normal circumstances?

from lffd-a-light-and-fast-face-detector-for-edge-devices.

@YonghaoHe

from lffd-a-light-and-fast-face-detector-for-edge-devices.

@huanpig indeed, the training speed is not fast. The data preparation is slow. The training speed is about 50 images/s.

from lffd-a-light-and-fast-face-detector-for-edge-devices.