2017-06-26 18:35:51 +08:00
Load Caffe framework models {#tutorial_dnn_googlenet}
===========================
2020-05-20 06:59:28 +08:00
@next_tutorial {tutorial_dnn_halide}
2017-06-26 18:35:51 +08:00
Introduction
------------
In this tutorial you will learn how to use opencv_dnn module for image classification by using
GoogLeNet trained network from [Caffe model zoo ](http://caffe.berkeleyvision.org/model_zoo.html ).
We will demonstrate results of this example on the following picture.
Source Code
-----------
2018-06-01 00:30:56 +08:00
We will be using snippets from the example application, that can be downloaded [here ](https://github.com/opencv/opencv/blob/3.4/samples/dnn/classification.cpp ).
2017-06-26 18:35:51 +08:00
2018-03-04 00:29:37 +08:00
@include dnn/classification.cpp
2017-06-26 18:35:51 +08:00
Explanation
-----------
-# Firstly, download GoogLeNet model files:
2018-03-04 00:29:37 +08:00
[bvlc_googlenet.prototxt ](https://github.com/opencv/opencv_extra/blob/master/testdata/dnn/bvlc_googlenet.prototxt ) and
2017-06-26 18:35:51 +08:00
[bvlc_googlenet.caffemodel ](http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel )
Also you need file with names of [ILSVRC2012 ](http://image-net.org/challenges/LSVRC/2012/browse-synsets ) classes:
2019-03-29 03:31:38 +08:00
[classification_classes_ILSVRC2012.txt ](https://github.com/opencv/opencv/blob/3.4/samples/data/dnn/classification_classes_ILSVRC2012.txt ).
2017-06-26 18:35:51 +08:00
Put these files into working dir of this program example.
-# Read and initialize network using path to .prototxt and .caffemodel files
2018-03-04 00:29:37 +08:00
@snippet dnn/classification.cpp Read and initialize network
2017-06-26 18:35:51 +08:00
2018-03-04 00:29:37 +08:00
You can skip an argument `framework` if one of the files `model` or `config` has an
extension `.caffemodel` or `.prototxt` .
This way function cv::dnn::readNet can automatically detects a model's format.
2017-06-26 18:35:51 +08:00
-# Read input image and convert to the blob, acceptable by GoogleNet
2018-03-04 00:29:37 +08:00
@snippet dnn/classification.cpp Open a video file or an image file or a camera stream
2017-06-26 18:35:51 +08:00
2018-03-04 00:29:37 +08:00
cv::VideoCapture can load both images and videos.
@snippet dnn/classification.cpp Create a 4D blob from a frame
We convert the image to a 4-dimensional blob (so-called batch) with `1x3x224x224` shape
after applying necessary pre-processing like resizing and mean subtraction
`(-104, -117, -123)` for each blue, green and red channels correspondingly using cv::dnn::blobFromImage function.
2017-06-26 18:35:51 +08:00
2018-03-04 00:29:37 +08:00
-# Pass the blob to the network
@snippet dnn/classification.cpp Set input blob
2017-06-26 18:35:51 +08:00
-# Make forward pass
2018-03-04 00:29:37 +08:00
@snippet dnn/classification.cpp Make forward pass
During the forward pass output of each network layer is computed, but in this example we need output from the last layer only.
2017-06-26 18:35:51 +08:00
-# Determine the best class
2018-03-04 00:29:37 +08:00
@snippet dnn/classification.cpp Get a class with a highest score
We put the output of network, which contain probabilities for each of 1000 ILSVRC2012 image classes, to the `prob` blob.
And find the index of element with maximal value in this one. This index corresponds to the class of the image.
-# Run an example from command line
@code
./example_dnn_classification --model=bvlc_googlenet.caffemodel --config=bvlc_googlenet.prototxt --width=224 --height=224 --classes=classification_classes_ILSVRC2012.txt --input=space_shuttle.jpg --mean="104 117 123"
@endcode
For our image we get prediction of class `space shuttle` with more than 99% sureness.
