Novel Object Detection

Novel Object Detection is a challenging computer vision task proposed by Fomenko et al. in their paper “Learning to Discover and Detect Objects”. The task aims to evaluate the model's mAP performance on both known and unknown categories, where the known categories are the 80 classes from the COCO dataset, and the unknown categories are the remaining 1123 classes from the LVIS dataset. During training, the model can only learn from the annotations provided in the COCO dataset, but during evaluation and inference, it must classify and detect all categories present in the LVIS dataset. This task is of significant application value for enhancing the model’s generalization ability and its capability to recognize new objects.