ml-neuman icon indicating copy to clipboard operation
ml-neuman copied to clipboard

Can I scale the image to accelerate the training?

Open buaacyw opened this issue 2 years ago • 2 comments

Thanks for your great work! I think downsampling the image is a good way to accelerate the training. But I haven't found any API in your codebase. Is there any? Thanks!

buaacyw avatar Oct 25 '22 07:10 buaacyw

Simply config the image_width and image_height raise a bug: Telling that fused_depth_map doesn't exist. So I copy this property from NeuManCapture to ResizedNeuManCapture. And then it trains smoothly. Since I haven't read the full code. Is this modification OK? Thanks!

buaacyw avatar Oct 25 '22 08:10 buaacyw

Yes, I think you can do that. If you can share some visualization of the resized fused depth that will be great.

jiangwei221 avatar Oct 25 '22 14:10 jiangwei221