PaperReading

TNT: Target-driveN Trajectory Prediction

February 2022

tl;dr: Predict diverse set of future targets and then use target to drive trajectory prediction.

Overall impression

The paper described the core drawbacks of previous methods, involving sampling latent states (VAE, GAN), or fixed anchors (coverNet, MultiPath).

TNT has the following advantages

The target, or final state capture most uncertainty of a trajectory. TNT decompose the distribution of futures by conditioning on targets, and then marginalizing over them.

The anchor-based method is improved by DenseTNT to be anchor-free, which also eliminated the NMS process by learning.

Key ideas

Technical details

Notes