夜间感知新突破！北航等提出红外主导的高效目标检测方案 | ECCV'26
Merged summary
TL;DR - InfraNet is an IR-centric object detector that uses quality-aware RGB guidance to improve robustness when visible-light imagery degrades. It supports both RGB-IR inference and efficient IR-only deployment.
- QualGate suppresses unreliable RGB features while amplifying infrared features to prevent negative transfer.
- InfraNet-IR uses RGB during training but removes the RGB branch and fusion modules for IR-only inference.
- The method achieved competitive results across LLVIP, FLIR-Aligned, M3FD, and DroneVehicle benchmarks.
- The design targets robust detection in nighttime, adverse-weather, surveillance, autonomous-driving, and drone scenarios.
Sources (1)
夜间感知新突破！北航等提出红外主导的高效目标检测方案 | ECCV'26
TL;DR - InfraNet is an IR-centric object detector that uses quality-aware RGB guidance to improve robustness when visible-light imagery degrades. It supports both RGB-IR inference and efficient IR-only deployment.
- QualGate suppresses unreliable RGB features while amplifying infrared features to prevent negative transfer.
- InfraNet-IR uses RGB during training but removes the RGB branch and fusion modules for IR-only inference.
- The method achieved competitive results across LLVIP, FLIR-Aligned, M3FD, and DroneVehicle benchmarks.
- The design targets robust detection in nighttime, adverse-weather, surveillance, autonomous-driving, and drone scenarios.