schedule_3x.py

# optimizer
# This schedule is mainly used by models on indoor dataset,
# e.g., VoteNet on SUNRGBD and ScanNet
lr = 0.008  # max learning rate
optimizer = dict(type='AdamW', lr=lr, weight_decay=0.01)
optimizer_config = dict(grad_clip=dict(max_norm=10, norm_type=2))
lr_config = dict(policy='step', warmup=None, step=[24, 32])
# runtime settings
total_epochs = 36