Change the size of inputs

shirakawas / ASNG-NAS

Adaptive Stochastic Natural Gradient Method for One-Shot Neural Architecture Search

MIT License

88 stars 11 forks source link

Change the size of inputs #2

Open 441711335 opened 4 years ago

441711335 commented 4 years ago

Hi @shirakawas , I changed the resolution of the input image to 100 when using main_inpainting_cat.py. Then I encountered a problem that the output resolution of the convolutional layer and the deconvolution feature resolution did not match. Could you offer me some suggestion? Thanks!

shirakawas commented 4 years ago

The current code only supports the image size of the power of 2 (2^n). To accept other image sizes, you need to modify the code for convolution and deconvolution layers and equalize the image sizes of corresponding layers.