Day 26/30: General Gaussian Heatmap Labeling is Arbitrary-Oriented Object Detection AOOD
π It uses an anchor-free object-adaptation label assignment strategy to deο¬ne positive candidates based on 2D GH, reο¬ecting shape and direction features of arbitrary-oriented objects.
π GGHL improves the AOOD performance with low parameter-tuning and time costs.
π It is also applicable to most AOOD methods to improve their performance including lightweight models.
π° Paper: A General Gaussian Heatmap Labeling for Arbitrary-Oriented Object Detection
Day 24/30: π₯ EfficientDet is a very popular object detection model for a good reason!
Letβs see why
π EfficientDet achieved State-Of-The-Art (SOTA) accuracy while reducing both the size of parameters, and the FLOPS, when it was released. Itβs still a very good contender.
π Before introducing EfficientDet, models were getting impressively big to achieve SOTA results
β The authors asked the following question:
Is it possible to build a scalable detection architecture with both higher accuracy and better efficiency across # resource constraints?
So, they systematically studied neural network architecture design choices for object detection, and proposed several key optimizations to improve efficiency:
1- A weighted bi-directional feature pyramid network (BiFPN), which allows easy and fast multiscale feature fusion
It was one of first competitors of anchor-based single/two stage object detectors.
Understanding FCOS will help understanding other model inspired by FCOS.
Summary ...π
π FCOS reformulates object detection in a per-pixel prediction fashion
π It uses multi-level prediction to improve the recall and resolve the ambiguity resulted from overlapped bounding boxes
π It proposes βcenter-nessβ branch, which helps suppress the low-quality detected bounding boxes and improves the overall performance by a large margin
π It avoids complex computation such as the intersection-over-union (IoU)