模型贡献至Open model zoo(3)
3.mobilemnet-yolo-syg.md
文件作用
说明文档，作用同README
(1)Use Case and High-Level Description
与model.yml的description相同
(2)Conversion（未用）
检测：命令参考
保存模型的目录名：model-mobilenet-yolo
下载权重：
wget -O weights/mobilenet-yolo-syg.h5 XXXX
-O：下载并以指定的文件名保存
21-34下载转换所需补丁并转换
模型转换成TF：
python3 keras_to_tensorflow-master/keras_to_tensorflow.py --input_model=“weights/mobilenet-yolo-syg.h5” –
output_model=“weights/mobilenet-yolo-syg.pb”
(3)Specification
| Metric | Value |
|---|---|
| Type | Detection |
| GFLOPs | 65.984 |
| MParams | 61.922 |
| Source framework | Keras* |
GFLOPs：24525195
MParams：12181862
Keras|TF计算模型的MParams参数量
操作
在train.py中主函数中
计算参数量
计算参数量
操作
修改yolo.py后运行yolo_image.py出现报告结束后ctrl+C
opts1 = tf.profiler.ProfileOptionBuilder.float_operation()# We use the Keras session graph in the call to the profiler.
FLOPs = tf.profiler.profile(graph=K.get_session().graph,
run_meta=run_meta, cmd=‘op’, options=opts1)
opts2 =
tf.profiler.ProfileOptionBuilder.trainable_variables_parameter()Params =
tf.profiler.profile(graph=K.get_session().graph,
run_meta=run_meta, cmd=‘op’, options=opts2)Y
print(“model FLOPs result is as follow:…”)print(FLOPs.total_float_ops)# Prints the “FLOPs” of the model.
print(“model Params result is as follow:…”)print(Params.total_parameters)# Prints the “Params” of the model
3.Accuracy
| Metric | Value |
|---|---|
| mAP | 35.38% |
| COCO* mAP (0.5) | 75.36% |
通过accuracy chcker计算所得
4.Input
Original model
B- batch sizeH- heightW- widthC- channel
Channel order isRGB.
Scale value - 255.
输入图像通道的确定
Converted model
B- batch sizeH- heightW- widthC- channel
Channel order isBGR.
5.Output
Original model
method1
netron查看pb文件查找
result
name - separable_conv2d_22, shape - 1,52,52,42
12,16, 19,36, 40,28
name - separable_conv2d_30, shape - 1,26,26,42
36,75, 76,55, 72,146
name - separable_conv2d_38, shape - 1,13,13,42
3个输出层名字，同accuracy-checkl.yml中outputs一样。
shape和原模型pb输出一样，但注意调整channel顺序。
detection box的形式参考yolov3和模型代码
name - separable_conv2d_30/BiasAdd/YoloRegion,
shape - 1,42,26,26. The anchor values are 36,75, 76,55, 72,146.
separable_conv2d_38/BiasAdd/YoloRegion，
shape - 1,42,13,13. The anchor values are 142,110, 192,243, 459,401
6.Legal Information
The original model is distributed under the following
license1:
license2: