United We Learn Better: Harvesting Learning Improvements From Class Hierarchies Across Tasks
This work addresses the need for hierarchical learning methods in object detection, enabling sigmoid-based architectures to benefit from class hierarchies, which is incremental as it extends existing classification approaches to a new task.
The paper tackles the problem of applying hierarchical learning improvements from classification to object detection, where existing methods are incompatible with sigmoid-based architectures, by introducing a theoretical framework for extracting parent predictions and a hierarchical loss that works across tasks, achieving results on both classification and detection benchmarks.
Attempts of learning from hierarchical taxonomies in computer vision have been mostly focusing on image classification. Though ways of best harvesting learning improvements from hierarchies in classification are far from being solved, there is a need to target these problems in other vision tasks such as object detection. As progress on the classification side is often dependent on hierarchical cross-entropy losses, novel detection architectures using sigmoid as an output function instead of softmax cannot easily apply these advances, requiring novel methods in detection. In this work we establish a theoretical framework based on probability and set theory for extracting parent predictions and a hierarchical loss that can be used across tasks, showing results across classification and detection benchmarks and opening up the possibility of hierarchical learning for sigmoid-based detection architectures.