search

Nightmare-n / GD-MAE

GD-MAE: Generative Decoder for MAE Pre-training on LiDAR Point Clouds (CVPR 2023)
Apache License 2.0
114 stars 6 forks source link

about separately and end-to-end #3

Closed phoebedddd closed 1 year ago

phoebedddd commented 1 year ago

What's the difference between the two training?

image

Nightmare-n commented 1 year ago

separately: when training a two-stage model, the weights of the first stage are frozen, so we need to train a single-stage model first, and then a two-stage model. end-to-end: we can directly train the two-stage model end-to-end (comment out the FREEZE_LAYERS in config files).

phoebedddd commented 1 year ago

separately: when training a two-stage model, the weights of the first stage are frozen, so we need to train a single-stage model first, and then a two-stage model. end-to-end: we can directly train the two-stage model end-to-end (comment out the FREEZE_LAYERS in config files).

Ok, I see. Thank you very much! Best wishes for you!