HRCenterNet

Multi-Scale Feature Representations

  We propose a generalized bird detection model, namely HRCenterNet with high accuracy inspired by HRNet and CenterNet. For feature extraction, the parallelized model is implemented so that the global feature and local feature can be considered together, and high-level and low-level representations can interact with each other under multi-scale fusion. In the final segmentation result, unlike anchor-based models in the past, our model can achieve scale-aware and accurate segmentation.