はてなブックマーク

Region of Interest pooling (RoI pooling) is a widely used technique in object detection with convolutional neural networks. It extracts fixed-size feature maps (e.g., 7×7) from variable-sized regions, enabling efficient classification of multiple objects like cars or pedestrians. RoI Align refines this process by avoiding coordinate quantization and using bilinear interpolation, resulting in more

We resized and aligned 16-band channels to match those from 3-band channels. Alignment was necessary to remove shifts between channels. Finally all channels were concatenated into single 20-channels input image. Model Our fully convolutional model was inspired by the family of U-Net architectures, where low-level feature maps are combined with higher-level ones, which enables precise localization.

The challenge The goal of the competition was to recognize individual right whales in photographs taken during aerial surveys. When visualizing the scenario, do not forget that these giants grow up to more than 18 metres, and weigh up to 91 tons. There were 447 different right whales in the data set (and this is likely the overall number of living specimens). Even though that number is (terrifying