Joint multi-configuration training followed by fine-tuning only the backbone and neck beats naive joint training on a new RoboTaxi/RoboBus 3D detection benchmark, by 1 to 2 mAP points.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Domain Adaptation for Different Sensor Configurations in 3D Object Detection
Joint multi-configuration training followed by fine-tuning only the backbone and neck beats naive joint training on a new RoboTaxi/RoboBus 3D detection benchmark, by 1 to 2 mAP points.