Asymmetric Loss For Multi-Label Classification

By Emanuel Ben Baruch, Tal Ridnik, Nadav Zamir, Asaf Noy, Itamar Friedman, Matan Protter, Lihi Zelnik-Manor. IEEE International Conference on Computer Vision, 2020

870 citations recorded by the source corpus.

In a typical multi-label setting, a picture contains on average few positive labels, and many negative ones. This positive-negative imbalance dominates the optimization process, and can lead to under-emphasizing gradients from positive labels during training, resulting in poor accuracy. In this paper, we introduce a novel asymmetric loss ("ASL"), which operates differently on positive and negative samples. The loss enables to dynamically down-weights and hard-thresholds easy negative samples, while also discarding possibly mislabeled samples. We demonstrate how ASL can balance the probabilities of different samples, and how this balancing is translated to better mAP scores. With ASL, we reach state-of-the-art results on multiple popular multi-label datasets: MS-COCO, Pascal-VOC, NUS-WIDE and Open Images. We also demonstrate ASL applicability for other tasks, such as single-label classification and object detection. ASL is effective, easy to implement, and does not increase the training time or complexity. Implementation is available at: https://github.com/Alibaba-MIIL/ASL.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/34ebd21d136dfcc6c6bee14ae142cc90f6b68110

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Lijuan Wang
2. Zicheng Liu
3. Jianfeng Wang
4. Zhe Gan
5. Xiaowei Hu
6. Tal Ridnik
7. Asaf Noy
8. Jianfeng Gao

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Dacheng Tao
3. Ming-Hsuan Yang
4. Bernt Schiele
5. Trevor Darrell
6. Xiaogang Wang
7. Jitendra Malik
8. Zhe Gan

Citations in the corpus

• AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (2022, methodology): “Since some images are not available on Flicker, we only use 220,000 images following [7, 32].”
• ImageNet-21K Pretraining for the Masses (2021, methodology): “As commonly done in multi-label classification [3], we reduce the problem to a series of binary classification tasks.”
• Injecting Semantic Concepts into End-to-End Image Captioning (2021, methodology): “Due to the extremely imbalanced semantic concepts distribution (certain concepts appear much frequently than the rest), we adopt the simplified asymmetric focal loss [6,43,44] which shows great performances handling sample imbalance problems for the multi-label classification task.”
• Recognize Anything: A Strong Image Tagging Model (2023, positive): “The tags parsed from the text are used as ground truth labels, and Asymmetric Loss [23] is employed for optimization.”
• CIFDM: Continual and Interactive Feature Distillation for Multi-Label Stream Learning (2021, methodology): “Therefore, the Asymmetric Loss[3] will be used here:”
• Re-labeling ImageNet: from Single to Multi-Labels, from Global to Localized Labels (2021, methodology): “To obtain machine-generated label maps, we utilize a pre-trained TResNet-XL model [1] whose input size is 640 × 640 and mAP is 88 .”