It proposes the Balanced Binary Angular Margin (BBAM) loss. This loss function helps balance the "variance bias" between positive and negative samples by looking at the angle of features in a multi-label setup.
A more recent 2024 paper titled introduces a different "BBAM". BBAM.rar
It uses a trained object detector to find the "smallest area" of an image that makes the detector produce the same result, effectively creating a map that identifies the object within the box. It proposes the Balanced Binary Angular Margin (BBAM) loss
Published at NeurIPS 2024 (Conference on Neural Information Processing Systems). Access: Available through the NeurIPS Proceedings . 3. Note on ".rar" Files It uses a trained object detector to find
You can find the full PDF and supplementary materials on arXiv or through the CVF Open Access portal. 2. S²ML²-BBAM: Balanced Binary Angular Margin
Published at CVPR 2021 (Conference on Computer Vision and Pattern Recognition).
This paper deals with Semi-supervised Multi-label Learning (SSMLL) , which tries to train models when only a portion of the data has multiple labels.