Open GuideWsp opened 5 years ago
Thanks for your interest in our work.
The jupyter script contains everything and running it directly gives you the exact same results included in the script (so the figures pasted in the paper). I'd suggest you provide your testing image, the result and filter flow map so that I can help debug.
@aimerykong , thanks for sharing amazing awesome work . Just a very basic question i had . How is this different from - OpenCv In-painting - does this build upon a Nearest Neighbor algorithm ? https://docs.opencv.org/master/df/d3d/tutorial_py_inpainting.html
Hi, Rohit,
Thanks for your interest!
You are exactly right in relating our work to nearest neighbor!
Our understanding is that, the filter flow is doing something like fast nearest neighbor. Different from nearest neighbor, the model learns to use the pixels and assign weights to form a weighted sum as the target pixel. Moreover, by using multigrid (~multiscale), we are able to use pixels in a larger neighborhood (non-local if one would think).
Regards, Shu
On Mon, Apr 20, 2020 at 3:10 AM Rohit Dhankar notifications@github.com wrote:
@aimerykong https://github.com/aimerykong , thanks for sharing amazing awesome work . Just a very basic question i had . How is this different from - OpenCv In-painting - does this build upon a Nearest Neighbor algorithm ? https://docs.opencv.org/master/df/d3d/tutorial_py_inpainting.html
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/aimerykong/predictive-filter-flow/issues/1#issuecomment-616355700, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABRJSJF6XQGINEFQZDAMLQDRNPYQFANCNFSM4GIHP7TA .
Recently, My team try to produce your paper's result by the model you proved(epoch-445.paramOnly) and the code(task01_deblur.ipynb). But the results is not as good as your paper, I wonder if it's something wrong about you code or your model? If so, could you provide the right code? Thanks for your reply.