Answers are grounded in this paper and cite their sources.
RGB–thermal detection benefits from both cross-modal feature calibration and the recovery of fine-scale structural evidence. However, spatial responses produced during these operations are often used only to regulate intermediate features. We propose a stage-specific semantic–structural descriptor-conditioning framework that retains calibration and discrepancy maps from P4 and structural-weighting and propagation maps from P2→P3. At P3–P5, the four maps are jointly encoded with pyramid features to generate separate, bounded adjustments for classification and regression. On FLIR Aligned, the framework achieves 46.83±0.10% mAP50:95 across six seeds, compared with 43.80±0.16% for the matched baseline (paired gain 3.03±0.15 points), using 7.51M parameters and 25.96 GFLOPs. It also outperforms a compute-matched ordinary P2 control by 1.72±0.10 points. Structural recovery provides the largest standalone gain. Under matched feature-path and prediction-head architectures, a six-seed four-way descriptor-group control shows a modest positive semantic–structural interaction (+0.066±0.031 points). Live descriptors exceed dual-neutral and cross-sample-permuted controls by 0.43±0.05 and 0.57±0.10 points. Independent retraining on M3FD yields a 3.17-point mAP50:95 gain. Stress tests further characterize sensitivity to spatially varying thermal misalignment.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: