scaling up strategy
Wolfwjs opened this issue · 1 comments
Wolfwjs commented
Thank you so much for such a great job! I have noticed from your paper that scaling up has played a very important role.
May I ask what are the model parameters of SparseConv & PointBERT after scaling up and what is the scaling strategy in detail ？
Colin97 commented
Hi, thanks for your attention. Please refer to the supplementary materials for the details of scaling strategy (section 6.4).
For the model parameters of SparseConv & PointBERT, please check our code. Here are the commands for the two backbones:
python3 src/main.py --trail_name spconv_all
python3 src/main.py --trail_name pointbert_all model.name=PointBERT model.scaling=4 model.use_dense=True training.lr=0.0005 training.lr_decay_rate=0.967