手法を比較

	弱教師あり物体検出	インスタンスセグメンテーション
分野	深層学習	深層学習
系統	Machine learning	Machine learning
提唱年	2016 (deep WSOD); MIL roots circa 1997	2017
提唱者	Bilen, H. & Vedaldi, A. (WSDDN); Multiple Instance Learning origins: Dietterich et al. (1997)	He, K., Gkioxari, G., Dollar, P., Girshick, R.
種類	Weakly supervised detection paradigm	Pixel-level detection and mask prediction
原典	Bilen, H., & Vedaldi, A. (2016). Weakly supervised deep detection networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2846–2854.	He, K., Gkioxari, G., Dollar, P., & Girshick, R. (2017). Mask R-CNN. Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2961–2969.
別名	WSOD, weakly-supervised detection, image-level supervised detection, multiple instance detection	instance-level segmentation, object instance segmentation, mask prediction, panoptic instance segmentation
関連	5	4
概要	Weakly Supervised Object Detection (WSOD) trains object detectors using only image-level labels — indicating which object classes appear in an image — without requiring costly bounding-box annotations. Multiple Instance Learning (MIL) formulations allow the model to discover the likely location of each object class from classification signals alone, dramatically reducing annotation cost.	Instance segmentation is a computer vision task that simultaneously detects every distinct object in an image and produces a precise pixel-level mask for each individual object instance. Unlike semantic segmentation, which labels every pixel with a class, instance segmentation distinguishes between separate objects of the same class, enabling fine-grained spatial understanding.