(YOLO的訓練模型框架)
git clone https://github.com/thtrieu/darkflow.git
python setup.py build_ext --inplace
須要安裝MS visual studio才能編譯安裝
------------------------------------------------d
Flowing the graph using
flowpython flow --help
Usage: 使用1顆GPU來run darkflow
python flow --model cfg/yolo.cfg --load bin/yolo.weights --gpu 1.0
All input images from default folder
sample_img/ are flowed through the net and predictions are put in sample_img/out/. python flow --imgdir sample_img/ --model cfg/yolo.cfg --load bin/yolo.weights --gpu 1.0
--json
json output can be generated with descriptions of the pixel location of each bounding box and the pixel location.
Training new model
flow --train will train the model . To point to training set and annotations, use option--dataset and --annotation.
1. ) 用既有的yolo network model (.cfg) ==> 所以你會從cfg/ copy 一份新的cfg並稍微修改一下檔名. 新的cfg檔, 要改2個地方
...
[convolutional]
size=1
stride=1
pad=1
filters=40 #在倒數第2個 [convolutional], 修改filters 數量
activation=linear
[region]
anchors = 1.08,1.19, 3.42,4.41, 6.63,11.38, 9.42,5.11, 16.62,10.52
bias_match=1
classes=3 #偵測的物件有3類
coords=4
num=5 # anchors box的數量 (一個anchors會有confidence、座標、在每個類別的機率)
softmax=1
...
2.) ---load xxxx. weights. ==> 會載入xxxx.weights時,會去 cfg/找到對應的xxxx.cfg ,如此才能載入權重, 如--load tiny-yolo-voc.weights 會去找 cfg/tiny-yolo-voc.cfg
When darkflow sees you are loading
tiny-yolo-voc.weights it will look for tiny-yolo-voc.cfg in your cfg/ folder and compare that configuration file to the new one you have set with --model cfg/tiny-yolo-voc-3c.cfg. In this case, every layer will have the same exact number of weights except for the last two, so it will load the weights into all layers up to the last two because they now contain different number of weights.3.) 修改 darkflow/labels.txt, 行數要和class數量一樣
aeroplane
bicycle
bird
例如. 若使用既有的tiny-yolo network model與weight
python flow --model cfg/tiny-yolo-joseph.cfg --load bin/tiny-yolo.weights --train --annotation train/annotations --dataset train/images --gpu 1.0
LabelImg 用來產生label 及框選Object
References:
https://github.com/thtrieu/darkflow
