EdwardSmith1884 / GEOMetrics

Repo for the paper "GEOMetrics: Exploiting Geometric Structure for Graph-Encoded Objects"
MIT License
117 stars 12 forks source link

Help #4

Closed shersoni610 closed 4 years ago

shersoni610 commented 4 years ago

Hello Edwards,

I am trying to understand your code. I am a newbie to python and Machine learning. I have a naive question. After training the data, how do I supply input dataset (i.e. image) to get a triangulated model?

EdwardSmith1884 commented 4 years ago

urr you could put it in a folder mimicking the data layout I already have. You would have to know the camera parameters of the image however.

KnightOfTheMoonlight commented 4 years ago

Is the final result based on epoch 3000? From your code https://github.com/EdwardSmith1884/GEOMetrics/blob/09471d70f441482319543bb5214eaaab38f64507/GEOMetrics.py#L375, I found there are 3000 epochs. And I didn't find the experiment details in your ICML2019 paper.

EdwardSmith1884 commented 4 years ago

no, I check the validation error regularly and stop training when the f1 score stops increasing.