Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The authors develop Boundary Feature Repair (BFR), a frozen-host adaptation framework that injects zero-initialized, task-supervised residual corrections directly at pre-fusion modality interfaces without modifying the underlying detector or its normalization statistics. This strategy bypasses the need to retrain large multimodal perception systems or rely on online adaptation, temporal cues, and corruption metadata during inference. On the 20-corruption KITTI-C benchmark, the combined camera-LiDAR repair module (BFR-CL) reduces Relative Corruption Error on Focals Conv-F from 14.29 to 10.46 and elevates corrupted AP from 73.65 to 77.01 while fully retaining clean performance.
Multimodal 3D detectors do not need costly end-to-end retraining to survive corrupted sensor inputs—inserting lightweight, zero-initialized residual adapters right at the pre-fusion boundary slashes corruption error while leaving the entire host architecture frozen.
Camera-LiDAR detectors can continue to consume unreliable features even when both sensors remain present, synchronized, and calibrated. We introduce \emph{Boundary Feature Repair} (BFR), a frozen-host adaptation framework that learns task-supervised residual corrections at modality interfaces the detector already consumes. BFR-C repairs each camera feature level read by fusion, whereas BFR-L aligns host-conditioned LiDAR candidates to a selected boundary and routes site-wise innovations relative to the frozen anchor. Their jointly trained composition is BFR-CL. Zero-initialized per-channel scales make every variant an exact detector-level identity before optimization; only the repair modules train, while the encoders, fusion consumer, router, detection head, and host normalization statistics remain fixed. At inference, BFR requires neither clean references, corruption metadata, temporal history, nor online updates. Across the complete 20-corruption, five-severity KITTI-C grid, BFR-C reduces RCE from $14.07$ to $11.92$ on MVX-Net and from $14.29$ to $11.00$ on Focals Conv-F relative to their reproduced frozen baselines. On the latter host, BFR-L raises AP$_{\mathrm{cor}}$ from $73.65$ to $74.48$, while BFR-CL reaches $77.01$ AP$_{\mathrm{cor}}$ and $10.46$ RCE with $86.02$ clean AP. On nuScenes-R, BFR-CL raises the reproduced MoME baseline's mAP robustness ratio from $80.1$ to $81.4$. These results establish boundary repair as a targeted retrofit for corrupted-but-present sensing without retraining the deployed detector.