ECCV 2026 | 北航提出InfraNet：红外目标检测新突破
Ranking
Overall
56
Content
65
Popularity
34
Observed public metrics from 1 member.
Merged summary
TL;DR - InfraNet uses quality-aware RGB guidance during training to improve infrared object detection, then removes the RGB branch for IR-only inference. This yields stronger robustness without increasing deployed model size or latency.
- QualGate suppresses unreliable RGB features while reinforcing multiscale IR representations.
- Asymmetric losses keep detection responsibility centered on the IR branch during training.
- On M3FD, InfraNet-IR improved AP from 57.2 to 59.0 while retaining 28.2M parameters and 12.5 ms inference.
- Experiments span LLVIP, FLIR-Aligned, M3FD, and DroneVehicle datasets.
Sources (1)
ECCV 2026 | 北航提出InfraNet：红外目标检测新突破
Public signals
Semantic Scholar citations 0 · Semantic Scholar influential citations 0
TL;DR - InfraNet uses quality-aware RGB guidance during training to improve infrared object detection, then removes the RGB branch for IR-only inference. This yields stronger robustness without increasing deployed model size or latency.
- QualGate suppresses unreliable RGB features while reinforcing multiscale IR representations.
- Asymmetric losses keep detection responsibility centered on the IR branch during training.
- On M3FD, InfraNet-IR improved AP from 57.2 to 59.0 while retaining 28.2M parameters and 12.5 ms inference.
- Experiments span LLVIP, FLIR-Aligned, M3FD, and DroneVehicle datasets.