(Review article on imbalance problems: Imbalance Problems in object Detection: A Review)
The imbalance problems we think of most are: target category imbalance, but in fact there are the following imbalance problems:
(1) category imbalance: foreground and background Unbalanced, the number of different types of input bounding boxes in the foreground is unbalanced;
(2) Scale imbalance: the scale of the input image and the bounding box is unbalanced, and the contribution of different feature layers to the final result is unbalanced;
(3) Spatial imbalance: the contribution of different sample return loss imbalance, uneven distribution of positive samples IoU, the target position in the image imbalance;
(4) the objective function imbalance: different tasks (such as regression and classification) contribution to global imbalances loss
attached Reference URL on github: (https://github.com/kemaloksuz/ObjectDetectionImbalance)
