search

charlesq34 / pointnet2

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space
Other
3k stars 889 forks source link

GTX1080Ti Training is taking too long to test this repo #106

Open NicksonYap opened 5 years ago

NicksonYap commented 5 years ago

@charlesq34

Hey Charles, and All,

Is there a possiblity to download pre-trained files? Is it too large or something?

I had difficulties understanding how to run PointNet v1 / v2 Right now I'm trying to train more/all models / run all python files hoping to see some visualizations or output obj But the training process takes too long

And Idea?

Thanks

TianyangChen357 commented 5 years ago

Hi, You can change the default max_epoch in each train.py to get a shorter time (couple hours). I use 100 for classification training, 30 for segmentation training, and 30 for semantic scene parsing. Moreover, the results are (average mean loss, mean accuracy), (0.35,0.90), (0.26,0.93) for the first two steps. I was running on the third one so far.