search

harveyslash / Deep-Image-Analogy-PyTorch

Visual Attribute Transfer through Deep Image Analogy in PyTorch!
MIT License
112 stars 25 forks source link

Dimension size OF test Image #1

Open karanchahal opened 6 years ago

karanchahal commented 6 years ago

Is your model multi scale ? And if so how does that work exactly ? Any helpful links for that would be very much appreciated

harveyslash commented 6 years ago

At the moment, it only supports 224 x 224 , but its fairly easy to allow multi scale. The Utils function (to tensor), the model's deconvolution , and the upscaling values will need to be changed in order for this to work.

neelkadia-zz commented 6 years ago

@harveyslash @karanchahal Have you guys tried this with 800x800 px Images? Of yes, can you please share code or open pull request?

karanchahal commented 6 years ago

Hey @harveyslash , you could try modifying your network architecture such that it supports a Feature Pyramid Network. That is much more efficient than feeding in resized images of various scales.