Is the dataset in neat the same as transfuser?

autonomousvision / neat

[ICCV'21] NEAT: Neural Attention Fields for End-to-End Autonomous Driving

MIT License

314 stars 47 forks source link

Is the dataset in neat the same as transfuser? #8

Closed exiawsh closed 2 years ago

exiawsh commented 2 years ago

Hello, Thanks for your wonderful work! I have doubt with the data generation, Is the dataset in this work the same as in transfuser? Shall I download the data in https://s3.eu-central-1.amazonaws.com/avg-projects/transfuser/data/14_weathers_minimal_data.zip and directly use it? And what's the data format for query_points? just simple (x,y,t)->(2,5,1) as in the paper described? If the semantic map range is huge, did you normalize the input data?

ap229997 commented 2 years ago

The dataset used for training models in this repo is different than the 14_weathers_minimal_data from the transfuser repo. The details regarding the dataset are provided in Sec. 2.5 of the supplementary.

The BEV semantics and waypoint prediction are always in the coordinate frame with the ego-vehicle at the bottom center and the query points are in (x,y,t) format.