Towards Few-Annotation Learning in Computer Vision: Application to Image Classification and Object Detection tasks