视觉编码器的结构选择

opendilab / LMDrive

[CVPR 2024] LMDrive: Closed-Loop End-to-End Driving with Large Language Models

Apache License 2.0

671 stars 58 forks source link

Closed CoderXuans closed 1 month ago

CoderXuans commented 1 month ago

@deepcs233 请问选择memfuser_baseline_return_feature或memfuser_baseline_e1d3_return_feature的依据是什么呢？这两个效果差不多吗？

deepcs233 commented 1 month ago

你好，效果差不多，但是e1d3模型更小，推理速度更快

CoderXuans commented 1 month ago

谢谢你的回复，现在我明白了