lucid
lucid copied to clipboard
tensorflow
new style transfer error (just recently started happening)
style transfer example requires content and style photo to be both same sized (800x,533) for example, and it requires they are in jpeg format with no transperency layer. following these rules, just recently the style transfer example no longe works. this is the error im receiving
Dimension 0 in both shapes must be equal, but are 800 and 533. Shapes are [800,533,3] and [533,800,3]. From merging shape 0 with other shapes. for 'stack_1' (op: 'Pack') with input shapes: [800,533,3], [533,800,3], [533,800,3].
I'm having exactly the same problem. AND if you get the dimensions of content and style the same e.g., 1000 x 1000 px this error during the optimization: TypeError: inner() got an unexpected keyword argument 'difference_to' I think some depreciated tf code kicked in? There were warnings.
found a temporary fix: launch a python 2.7 env using the following link below, than import the code into it and run, it than works https://colab.research.google.com/notebook#create=true&language=python2
here is my working example below based off my repository(https://github.com/400lbhacker/beast-pytorch-tensorflow-neural-style-transfer) https://colab.research.google.com/drive/1IgiqQ5dN5G2W2ntrNRtxOpeC1SY95GPn
I have resolved this issue, I can patch any lucid project one is working on if you email me: [email protected], I cannot post to github due to the nature of the way colab is begging to default on py3 but i have a private repository that is pre configured that i can share as an example. I will update my beast neural style repository (based off lucid) with this snippet of info soon. have wonderful day ^-^
public fix for everyone to use is here I apologize for lateness: github: https://github.com/400lbhacker/Lucid-tensorflow-2d-neural-style-WORKING-FINAL colab: https://colab.research.google.com/github/400lbhacker/Lucid-tensorflow-2d-neural-style-WORKING-FINAL/blob/main/Final_Style.ipynb