Papers › CAFF-DINO: Multi-spectral object detection transformers with cross-attention features fusion

CAFF-DINO: Multi-spectral object detection transformers with cross-attention features fusion

27 Sep 2024IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2024 9archive 2025-07-28

Kevin Helvig, Baptiste Abeloos, Pauline Trouve-Peloux

Object detection on images can find benefit from coupling multiple spectra, each presenting specific useful features. However, building an efficient architecture coupling the different modalities is a complex task. Transformers, due to their ability to extract meaningful correlations between the different regions of the inputs appear as a promising way to perform features fusion across different spectra. This work presents a multi-spectral object detection architecture based on cross-attention features fusion (CAFF), combined with a transformer based detector (DINO). We demonstrate here the performance of the proposed approach in object detection compared with state-of-the-art approaches, on infrared-visible multi-spectral datasets. Moreover the robustness to systematic misalignment between image pairs is studied. The proposed approach is generic to any mono-spectrum transformer based detectors. The model developed in this study will be available in a dedicated github repository.

PaperPDF

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Multispectral Object DetectionObjectObject DetectionPedestrian Detectionobject-detection

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Multispectral Object Detection FLIR CAFF-DINO mAP 50.5% #3 of 18 Archive leaderboard report
Multispectral Object Detection FLIR CAFF-DINO mAP50 85.5% #3 of 18 Archive leaderboard report
Pedestrian Detection LLVIP CAFF-DINO AP 0.685 #2 of 15 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections