关于voi模型的训练

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

Open Karkers opened 9 months ago

Karkers commented 9 months ago

你好，我想知道voi模型如何能让其端到端的训练，其image分支只能加载预训练模型么？

Nightmare-n commented 9 months ago

没试过端到端训练的效果，把配置文件的FREEZE_LAYERS: ['DLASeg']删掉应该就能端到端训练了