Computer Vision News - September 2018

Now that the space of shared parameters between the neurons is in a number of resolutions, the sharing of the image space is much more tightly aligned and therefore the movement of visual landmarks between the differently combined images we are studying will be largely reduced. And when we run the following code: We’ll get the following result: Images from https://distill.pub/ (CC-BY 4.0) 21 Tool Computer Vision News Focus on… def interpolate_f(): unique = spatial.fft_image((6, 128, 128, 3)) shared = [ lowres.lowres_tensor((6, 128, 128, 3), (1, 128//2, 128//2, 3)), lowres.lowres_tensor((6, 128, 128, 3), (1, 128//4, 128//4, 3)), lowres.lowres_tensor((6, 128, 128, 3), (1, 128//8, 128//8, 3)), lowres.lowres_tensor((6, 128, 128, 3), (2, 128//8, 128//8, 3)), lowres.lowres_tensor((6, 128, 128, 3), (1, 128//16, 128//16, 3)), lowres.lowres_tensor((6, 128, 128, 3), (2, 128//16, 128//16, 3)), ] return color.to_valid_rgb(unique + sum(shared), decorrelate=True) objective = objectives.channel_interpolate(*neuron1, *neuron2) images = render.render_vis(model, objective, param_f=interpolate_f, verbose=False) Read at page 24 our interview with Iro Armeni , our Woman in Computer Vision of September. Read also many more interviews with Women in Science on our online archive.