#siamese networks
4 resultsBag of Negatives for Siamese Architectures
Bojana Gajic, Ariel Amato, Ramon Baldrich +1
The paper introduces Bag of Negatives (BoN), a loss‑independent method that uses online hashing to efficiently select high‑quality negative samples for training Siamese networks on…
Tracking Holistic Object Representations
Axel Sauer, Elie Aljalbout, Sami Haddadin
The paper proposes a framework that augments existing siamese‑based visual trackers by dynamically storing a diverse set of object templates during tracking, improving robustness a…
Model Decay in Long-Term Tracking
Efstratios Gavves, Ran Tao, Deepak K. Gupta +1
The paper studies how updating a visual tracker with incorrect bounding boxes introduces a bias called model decay, which accumulates over long videos and causes tracking drift, an…
Few-Shot Viewpoint Estimation
Hung-Yu Tseng, Shalini De Mello, Jonathan Tremblay +4
The paper introduces a meta-learning framework called MetaView for category-level few-shot viewpoint estimation, enabling accurate pose prediction for new object categories using t…