Multi scan radar object classification on RadarScenes [P]

A new study on multi-scan radar object classification using RadarScenes shows significant improvements over a single-scan baseline. Point pooling across 20 scans increased the Macro F1 score by +0.1243 to 0.8613. A Causal GRU further improved the score by +0.0282 to 0.8895. Fusion with pooled embedding added a negligible +0.0002, indicating that the GRU provided most of the sequential modeling benefits beyond simple pooling.