Learning device, learning method, and image segmentation device

Abstract

The system includes a learning data acquisition unit (100) that acquires learning data which is a combination of a learning image, a correct edge image indicating a correct edge in the learning image, and correct geometric parameters related to the shape of the correct edge; an edge estimation unit (200) that is configured with a neural network and outputs an estimated edge image indicating an estimated edge of the learning image and estimated geometric parameters related to the shape of the estimated edge when the learning image is input to the neural network; a cost calculation unit (300) that uses the correct edge image, the correct geometric parameters, the estimated edge image, and the estimated geometric parameters to calculate a cost for evaluating the estimation accuracy by the edge estimation unit; and a model parameter update unit (400) that updates model parameters in the neural network using the cost calculated by the cost calculation unit.

Images