peterl1n / backgroundmattingv2-tensorflow Goto Github PK

Is TF implementation faster than torch one?

Hi! I'm having very strange outputs in TensorflowJS w.r.t. the ones I obtain running locally the MattingRefine model.

Everything is fine if the refine mode is "Full" or I use only the MattingBase network.

But if I use in TensorflowJS the "sampling" and "thresholding" refine modes within the MattingRefine model I obtain wrong results, like showed in the figures.
The images are simply the foreground output concatened with the alpha channel. Background and bottle are the same.

Tensorflow JS output (MattingRefine w/ thresholding mode)

Tensorflow output (MattingRefine w/ thresholding mode)

I think this is related to different implementations of operators between Keras and TensorflowJS.
Do you have any idea where are the instructions wrongly interpreted?

Thanks for your help and congrats for your great job!

Notes: I converted the Keras model first to Saved Model format and then to TFJS using TFJS-converter.

Hi， how to download the tensorflow version weights file？

https://drive.google.com/drive/folders/1cbetlrKREitIgjnIikG1HdM4x72FtgBh
Only Support onnx、torch、torchscripts weights file.
Thanks

Could you please expand the example snippet to a full example?
This would be really helpfull.
Thanks.