search

laughtervv / SGPN

SGPN:Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation, CVPR, 2018
MIT License
268 stars 62 forks source link

my dataset(cityscape) got : same: 1.000000, diff: 1.000000, pos: 0.000000 #14

Open WenshuangSong opened 6 years ago

WenshuangSong commented 6 years ago

Hi @laughtervv ,

I've been trying to run your code on the cityscape dataset. However I've got the log as follows: Batch: 19, loss: 7658.622900, grouperr: 0.190640, same: 1.000000, diff: 1.000000, pos: 0.000000 I don't know why ,could you please give me some advice?

Thanks a lot !

hailingluo commented 5 years ago

@laughtervv I have obtained the similar results with @songwenshuang (Both in my own datasets or the given dataset (Area_5_office_20.h5)). And I changed the margin too, but it doesn't work. Because of the memory of GPU, I couldn't enlarge the batch size. I'm looking forward for your reply. Thanks a lot.

densechen commented 5 years ago

I was trapped in the issue, too. I have solved this problem by set the parameter 'is_training = True' when I trained on my dataset, which is defined in the 'train.py->def train_one_epoch(epoch_num)'. @songwenshuang @Lareina-LUO

Xieshichen commented 5 years ago

Batch: 1719, loss: 227.894560, grouperr: 0.146339, same: 0.103193, diff: nan, pos: 0.210109 @LittleLampChen have you encountered the log in the trian.py. i dont know the reason that diff is nan. I'm looking forward for your reply. Thanks

NBSWw commented 1 year ago

@Xieshichen I have the same problem as you. Have you solved it? Look forward to your reply

ZhuangTingT commented 9 months ago

@Xieshichen I have the same problem as you. Have you solved it? Look forward to your reply

i find setting a large batch_size works. i set it as 12.