Cross-dataset Training for Class Increasing Object Detection
This addresses the time-consuming labeling issue in industrial applications where object classes increase on demand, though it is incremental as it builds on existing object detection frameworks.
The paper tackles the problem of detecting the union of object classes from multiple labeled datasets without requiring labeling of all classes across all datasets, achieving similar performance to independent training on datasets like PASCAL VOC, COCO, WIDER FACE, and WIDER Pedestrian.
We present a conceptually simple, flexible and general framework for cross-dataset training in object detection. Given two or more already labeled datasets that target for different object classes, cross-dataset training aims to detect the union of the different classes, so that we do not have to label all the classes for all the datasets. By cross-dataset training, existing datasets can be utilized to detect the merged object classes with a single model. Further more, in industrial applications, the object classes usually increase on demand. So when adding new classes, it is quite time-consuming if we label the new classes on all the existing datasets. While using cross-dataset training, we only need to label the new classes on the new dataset. We experiment on PASCAL VOC, COCO, WIDER FACE and WIDER Pedestrian with both solo and cross-dataset settings. Results show that our cross-dataset pipeline can achieve similar impressive performance simultaneously on these datasets compared with training independently.