用yolo检测动物 分5类 老虎 兔子 鱼 鹿 猫。训练的图片只有189张 迭代3000次的效果

./darknet yolo test cfg/yolo-tiny-animal.cfg backup/yolo-tiny-animal_final.weights

cfg/yolo-tiny-animal.cfg 文件内容

[net]batch 64subdivisions 64height 448width 448channels 3momentum 0.9decay 0.0005learning_rate 0.0001policy stepssteps 20,40,60,80,20000,30000scales 5,5,2,2,.1,.1max_batches 40000[crop]crop_width 448crop_height 448flip 0angle 0saturation 1.5exposure 1.5[convolutional]filters 16size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 32size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 64size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 128size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 256size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 512size 3stride 1pad 1activation leaky[maxpool]size 2stride 2[convolutional]filters 1024size 3stride 1pad 1activation leaky[convolutional]filters 1024size 3stride 1pad 1activation leaky[convolutional]filters 1024size 3stride 1pad 1activation leaky[connected]output 256activation linear[connected]output 4096activation leaky[dropout]probability .5[connected]output 735activation linear[detection]classes 5coords 4rescore 1side 7num 2softmax 0sqrt 1jitter .2object_scale 1noobject_scale .5class_scale 1coord_scale 5

做图像标签的脚本  darknet-animal/data/make_labels.py

# -*- coding: utf-8 -*- import os l [ rabbit , fish , deer , tiger , cat , 兔子 ]for word in l: os.system( convert -fill black -background white -bordercolor white -border 4 -font /usr/share/fonts/truetype/arphic/ukai.ttc -pointsize 18 label:\\ %s\\ \\ %s.png\\ %(word, word))