Meta Guided Metric Learner for Overcoming Class Confusion in Few-Shot Road Object Detection

Localization and recognition of less-occurring road objects have been a\nchallenge in autonomous driving applications due to the scarcity of data\nsamples. Few-Shot Object Detection techniques extend the knowledge from\nexisting base object classes to learn novel road objects given few training\nexamples. Popular techniques in FSOD adopt either meta or metric learning\ntechniques which are prone to class confusion and base class forgetting. In\nthis work, we introduce a novel Meta Guided Metric Learner (MGML) to overcome\nclass confusion in FSOD. We re-weight the features of the novel classes higher\nthan the base classes through a novel Squeeze and Excite module and encourage\nthe learning of truly discriminative class-specific features by applying an\nOrthogonality Constraint to the meta learner. Our method outperforms\nState-of-the-Art (SoTA) approaches in FSOD on the India Driving Dataset (IDD)\nby upto 11 mAP points while suffering from the least class confusion of 20%\ngiven only 10 examples of each novel road object. We further show similar\nimprovements on the few-shot splits of PASCAL VOC dataset where we outperform\nSoTA approaches by upto 5.8 mAP accross all splits.\n