本项目 Fork 自 https://github.com/ultralytics/yolov3 感谢大佬开源！！！
pip install -U -r requirements.txt以下以训练 VOC2007 数据集为例，tiny 预训练模型
VOC数据文件结构如下
- VOC2007
- Annotations # XML文件
- ImageSets
- Main
|- test.txt
|- train.txt
|- trainval.txt
|- val.txt
- JPEGImages # 图片 我们将 VOC 数据集下载下来放在项目的 data 文件夹中
cd data # 进入本项目的 data 文件夹
ln -s VOC2007/JPEGImages ./images # 建立软连接mkdir images
ln -s /input0/VOC2007/JPEGImages ./home/yolov3/data/images/
aeroplane
bicycle
bird
boat
bottle
bus
car
cat
chair
cow
diningtable
dog
horse
motorbike
person
pottedplant
sheep
sofa
train
tvmonitorclasses= 20
train=data/train_voc_2007.txt # 这个放在 VOC.zip 文件下了，解压后移动到 data 下面即可
valid=data/val_voc_2007.txt
names=data/custom.namescfg/yolov3-tiny-20cls.cfg
训练Training和测试 Testing需要修改一下
[net]
# Testing
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=2
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.001
burn_in=1000
max_batches = 500200
policy=steps
steps=400000,450000
scales=.1,.1
[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=1
[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky
###########
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=75
activation=linear
[yolo]
mask = 3,4,5
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=20
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
[route]
layers = -4
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[upsample]
stride=2
[route]
layers = -1, 8
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=75
activation=linear
[yolo]
mask = 0,1,2
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=20
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1tiny 网络
cd weights
wget -c https://pjreddie.com/media/files/yolov3-tiny.weightspython train.py --data data/custom.data --cfg cfg/yolov3-tiny-20cls.cfg --weights weights/yolov3-tiny.weights --epoch 100 --batch 64 --accum 1 --multiEpoch gpu_mem GIoU obj cls total targets img_size
9/272 7.08G 4.05 1.84 4.34 10.2 29 416: 100%|█████| 157/157 [01:37<00:00, 1.60it/s]
Class Images Targets P R [email protected] F1: 100%|█████| 79/79 [00:28<00:00, 2.73it/s]
all 2.51e+03 6.31e+03 0.39 0.674 0.429 0.487python detect.py --cfg cfg/yolov3-tiny-20cls.cfg --names data/custom.names --weights weights/best.ptpython test.py --cfg cfg/yolov3-tiny-20cls.cfg --data data/custom.data --weights weights/best.pt Class Images Targets P R [email protected] F1: 100%
all 2.51e+03 6.31e+03 0.0207 0.852 0.534 0.0404
aeroplane 2.51e+03 155 0.0349 0.903 0.681 0.0672
bicycle 2.51e+03 177 0.0158 0.921 0.324 0.031
bird 2.51e+03 243 0.0182 0.761 0.525 0.0355
boat 2.51e+03 150 0.0184 0.753 0.466 0.0359
bottle 2.51e+03 252 0.0131 0.504 0.146 0.0256
bus 2.51e+03 114 0.018 0.912 0.726 0.0353
car 2.51e+03 625 0.0241 0.789 0.608 0.0467
cat 2.51e+03 190 0.0379 0.926 0.752 0.0728
chair 2.51e+03 398 0.0204 0.847 0.408 0.0398
cow 2.51e+03 123 0.0169 0.943 0.5 0.0332
diningtable 2.51e+03 112 0.0219 0.902 0.565 0.0428
dog 2.51e+03 257 0.0233 0.946 0.646 0.0455
horse 2.51e+03 180 0.0183 0.983 0.799 0.0359
motorbike 2.51e+03 172 0.0179 0.942 0.348 0.0351
person 2.51e+03 2.33e+03 0.0186 0.916 0.518 0.0365
pottedplant 2.51e+03 266 0.0138 0.602 0.202 0.0269
sheep 2.51e+03 127 0.0131 0.717 0.46 0.0258
sofa 2.51e+03 124 0.0188 0.935 0.473 0.0368
train 2.51e+03 152 0.0381 0.974 0.894 0.0733
tvmonitor 2.51e+03 158 0.0136 0.873 0.642 0.0268
python plot_results.py 
以下的环境测试成功过
numpy 1.18.1
opencv-python 4.1.1.100
pandas 1.0.0
Pillow 6.2.2
scikit-image 0.16.2
scikit-learn 0.22.1
scipy 1.4.1
tensorboard 1.14.0
tensorflow 1.14.0.100
tensorflow-estimator 1.14.0
tensorflow-gpu 1.14.0
tensorflow-serving-api 1.14.0
torch 1.2.0
torchvision 0.4.0
tqdm 4.42.1 python train.py --data data/custom.data --cfg cfg/yolov3-tiny-20cls.cfg --img-size 416 --epochs 2 --evolve --weights weights/yolov3-tiny.weights 需要修改 train.py 最后面 for _ in range(1): # generations to evolve