search

hjweide / convnet-for-geometric-matching

A Lasagne and Theano implementation of the paper "Convolutional neural network architecture for geometric matching" by Ignacio Rocco, Relja Arandjelović, and Josef Sivic
MIT License
29 stars 9 forks source link

Demo code #2

Open fastlater opened 6 years ago

fastlater commented 6 years ago

Hi, I can see your results in the readme but do you have a demo code to visually these results? Like https://github.com/ignacio-rocco/cnngeometric_pytorch/blob/master/demo.py

I was following the original project and I would like to test your code and provide feedback if I find something that may interest you.

Thanks and advance.

hjweide commented 6 years ago

Yes, please see the warp_images.py file. You can specify pairs of filenames for warping. The results are the three images plotted together as one (like the example images).

Note that I haven't implemented the thin-plate-spline transform yet, so I don't expect my code to reproduce the results from the paper. If you find any bugs or other problems with my code, please let me know though :)

fastlater commented 6 years ago

@hjweide Thank you for your quick reply. I will check the warp_images.py. Actually, I am only interested on the affine results. I would like to reproduce the affine results using tensorflow and your code in theano is easier to read for me than pytorch. Let me know if you know any github where someone already did it in tensorflow so I can follow that project too.