面向密集预测任务的点云Transformer适配器

张德军,白燕子,曹锋,吴亦奇,徐战亚

Point cloud Transformer adapter for dense prediction task

Dejun ZHANG,Yanzi BAI,Feng CAO,Yiqi WU,Zhanya XU

表 6 加载各种预训练模型权重的定量比较

Tab.6 Quantitative comparison of loaded weights from various pretrained models

方法 预训练模型权重 mIoU/% mAcc/% 标准Transformer — 62.9 72.1 ACT[7] 63.1 73.2 Point-MAE[9] 63.1 72.1 MaskPoint[8] 64.4 72.7 Point-Bert[10] 63.5 75.7 ReCon[3] 64.8 73.3 PCT-Adapter — 66.2 73.7 ACT[7] 66.9 75.5 Point-MAE[9] 68.9 76.5 MaskPoint[8] 68.2 75.9 Point-Bert[10] 69.0 80.5 ReCon[3] 67.4 74.7