yolov4 模型转rknn模型，在rk3399prod推理一帧时间2.3s

[复制链接] · 发表于 2021-10-27 17:45:38

模型转换代码参考官方提供的yolov3_demo，转换代码如下

from PIL import Image
import numpy as np
#from matplotlib import pyplot as plt

import re
import math
import random

from rknn.api import RKNN

if __name__ == '__main__':

# Create RKNN object
rknn = RKNN()

# Load tensorflow model
print('--> Loading model')
rknn.load_darknet(model='./yolov4-custom.cfg', weight="./yolov4-custom_best.weights")

print('done')

rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2', batch_size=1)

# Build model
print('--> Building model')
rknn.build(do_quantization=True, dataset='./dataset_608x608.txt', pre_compile=True)
print('done')

rknn.export_rknn('./yolov4_person_hand_1_608x608.rknn')

exit(0)

转换后在板子上推理时间竟然要2.3s，yolov4原模型是250M,用darknet训练，转换后是63M。大家遇到过这样的问题吗？