Problem
Research questions and friction points this paper is trying to address.
tiny object detection
pollinator detection
class imbalance
small-scale objects
field video analysis
Where Is the Bee? Detecting Tiny Pollinators with a Single Collaborative-Head Transformer
🤖 AI Summary
This work addresses the challenging task of detecting extremely small pollinator insects in high-resolution field images, where target instances occupy on average only 0.16% of the image area and exhibit severe class imbalance (with bees comprising 80% of all annotations). To tackle these issues, the authors propose a detection framework built upon Co-DINO with a Swin-L backbone, enhanced by a crop-and-paste data augmentation strategy to increase the representation of rare classes and a class-weighted Simplex ETF loss to strengthen the discriminative capacity of query vectors. Without resorting to test-time augmentation or model ensembling, the proposed method substantially mitigates the dual challenges of extreme scale variation and category imbalance, achieving state-of-the-art performance with an mAP@[.5:.95] of 0.5062 and securing first place in the CVPPA@ECCV 2026 BuzzSpot Challenge FinalTest.
📝 Abstract
The CVPPA@ECCV 2026 BuzzSpot Challenge asks us to detect bees, bumblebees, hoverflies, and moths in 1920x1080 field keyframes. Its annotations carry 2 difficulties: the median box occupies 0.16% of a frame, and bees account for 80% of the labels. To cope with the small boxes, we compare 10 recorded detector configurations on held-out keyframes; plain Co-DINO with a Swin-L backbone has the highest mAP in this comparison, so we select it. Training then addresses the bee dominance in 2 ways: fine-tuning on a crop-mosaic pool in which the combined annotation share of the 3 rare classes rises from 19.9% to 55.1%, and a class-weighted simplex equiangular tight frame (ETF) loss that pulls the projected states of matched decoder queries toward fixed class directions. The full schedule spans 12+3+2 epochs. Without inference-time ensembling or test-time augmentation, we rank first on FinalTest at 0.5062 mAP@[.5:.95].
Innovation
Methods, ideas, or system contributions that make the work stand out.
Co-DINO
small object detection
class-imbalanced learning
ETF loss
crop-mosaic augmentation
🔎 Similar Papers
No similar papers found.
