Hi, yuan,
When I try to test main.cpp, I found that AutoWhiteBalance class does not comtain "Apply" function in main.cpp line 32. Is there something wrong with it?

ps, could you please share me your opencvConfig.cmake. When I cmake CMakeLists.txt, there is always something wrong with opencv.

Thank you very much.

I am trying to understand the "Convolutional Color Constancy" paper. Can you please comment and explain how is the log chroma histogram built and how to actually visualize it?
In the visualize_hist.m function, what is the input X and rho?
I only want to visualize the 2D log chroma histogram.

Hi, yuanxy92.
Thanks for your great implementation!
During my work, I have a question about Cpp implementation codes.
In Cpp/main.cpp Line 32 and 33, you did autoWB.apply() twice.
I commented one of them and found there was no difference in their results.
Do you have any reason to do so?

By the way, I think this function has typo: autoWB.apply() has to be autoWB.calc().
Thanks.

Hi Yuan,

Thanks for the great work! Could you give some instructions about extending the training data?
The pretrained model is not really optimized for my image sensor. I would like to train the system specifically for the images captured by my image sensor.

Thanks!
Burak

Hi yuan, thanks for your sharing.

I used your code to test my own images while the results look not good.
Is there any limitation about this method? PNG16 raw images are used for training. Does it work for normal 8bit image?

• input image:

• output image:

(I modified the funcion name in main.cpp. Is it a correct usage? )

int main(int argc, char* argv[]) {
	AutoWhiteBalance autoWB;
	autoWB.loadModel("E:/Project/AutoWhiteBalance/data/model/model.bin");

	cv::Mat img = cv::imread("E:/data/giga/NanshanIPark/2/calibrate/ref_00.jpg");
	cv::cuda::GpuMat img_d;
	img_d.upload(img);

	float gain_r, gain_g, gain_b;

	time_t begin, end;
	begin = clock();

	// autoWB.apply(img_d, gain_r, gain_g, gain_b);
	// autoWB.apply(img_d, gain_r, gain_g, gain_b);

        autoWB.calc(img_d, gain_r, gain_g, gain_b);

	end = clock();
	printf("Auto white balance update, cost %f milliseconds ...\n",
		static_cast<float>(end - begin) / static_cast<double>(CLOCKS_PER_SEC) * 1000);

	autoWB.applyWhiteBalance(img_d, gain_r, gain_g, gain_b);

	cv::Mat img2;
	img_d.download(img2);

	return 0;
}