Advancements in Efficient Annotation and Labeling for Computer Vision

The field of computer vision is moving towards more efficient and accurate annotation and labeling methods. Researchers are exploring ways to reduce the burden of manual annotation, which is time-consuming and expensive. One direction is the use of active learning and pseudo-labeling strategies to select the most informative samples for labeling and improve the accuracy of pseudo-labels. Another area of focus is the development of methods that can learn from partially annotated or noisy data, such as coarse annotations or label errors. These advancements have the potential to improve the performance of computer vision models and reduce the need for large amounts of annotated data. Noteworthy papers include:

  • Box-Level Class-Balanced Sampling for Active Object Detection, which proposes a class-balanced sampling strategy to improve the accuracy of pseudo-labels.
  • Learning to Detect Label Errors by Making Them, which presents a unified method for detecting label errors in object detection, semantic segmentation, and instance segmentation datasets.
  • POEv2: a flexible and robust framework for generic line segment detection and wireframe line segment detection, which achieves state-of-the-art performance on three publicly available datasets.

Sources

Box-Level Class-Balanced Sampling for Active Object Detection

Sketchpose: Learning to Segment Cells with Partial Annotations

Learning to Detect Label Errors by Making Them: A Method for Segmentation and Object Detection Datasets

Emerging Semantic Segmentation from Positive and Negative Coarse Label Learning

Estimating Conditional Covariance between labels for Multilabel Data

POEv2: a flexible and robust framework for generic line segment detection and wireframe line segment detection

More Reliable Pseudo-labels, Better Performance: A Generalized Approach to Single Positive Multi-label Learning

Built with on top of