Training time about pwc-net HOT 5 CLOSED

You can refer to Table 7 in our paper https://arxiv.org/pdf/1709.02371.pdf for the training time using a single NVIDIA Pascal TitanX GPU.

from pwc-net.

Thanks for the reply

from pwc-net.

I wonder if the training time mentioned in Table 7 includes all the dataset training(Chairs + Things + Sintel)? or just on FlyingChairs, thanks!

from pwc-net.

//.flo file to remap Mat //String addr: the .flo file name // cv::Mat flo: the Mat that can use function remap in opencv: cv::remap(img1,img2,flo,cv::Mat(),1); void floread(cv::String addr, cv::Mat &flo) { std::ifstream fin(addr, std::ios::binary); char buffer[sizeof(float)]; fin.read(buffer, sizeof(float)); float tar = ((float*)buffer)[0]; if (tar != TAR) { fin.close(); return; } fin.read(buffer, sizeof(int)); int width = ((int*)buffer)[0]; fin.read(buffer, sizeof(int)); int height = ((int*)buffer)[0]; //std::cout << width << "\t" << height << std::endl; flo = cv::Mat(cv::Size(width, height), CV_32FC2); for (int i = 0; i < height; i++) { for (int j = 0; j < width; j++) { if (!fin.eof()) { float * data = flo.ptr<float>(i, j); fin.read(buffer, sizeof(float)); data[0] = j + ((float*)buffer)[0]; fin.read(buffer, sizeof(float)); data[1] = i + ((float*)buffer)[0]; } } } fin.close(); }
the code transform from .flo file to remap Matrix,so that I get the remap image ,but it's not very good ,do you know why??

from pwc-net.

I wonder if the training time mentioned in Table 7 includes all the dataset training(Chairs + Things + Sintel)? or just on FlyingChairs, thanks!

Hi, Did you figure it out? I prefer that the training time including training in chairs + things. Because all the model of Sintel,kitti2012,kitti2015 are finetuned from it.

from pwc-net.