search

autonomousvision / neat

[ICCV'21] NEAT: Neural Attention Fields for End-to-End Autonomous Driving
MIT License
314 stars 47 forks source link

Effects of semantic #7

Closed Watson52 closed 2 years ago

Watson52 commented 2 years ago

Thanks for the work and the open source code. In the controller, I know that the use of red light for the longitudinal controller. But what if just use raw data input of RGB to identify instead of semantic? Could you please tell me the affect of the semantic? That is I think the semantic is just use to identify the traffic light.

ap229997 commented 2 years ago

Semantic loss is also crucial for training our model. You can see in Table 3 our supplementary material that there is a significant drop in performance when we remove the semantic loss.