The paper introduces a novel approach called Simultaneous Detection and Segmentation (SDS) to detect all instances of a category in an image and segment the pixels that belong to each instance. Unlike traditional bounding box detection or semantic segmentation, SDS requires both detection and precise pixel-level segmentation. The authors build on the Region Proposal Network (R-CNN) by using convolutional neural networks (CNNs) to classify category-independent region proposals and refine these proposals using category-specific figure-ground predictions. They achieve a 7-point boost in SDS performance, a 5-point boost in semantic segmentation, and state-of-the-art performance in object detection. The paper also provides diagnostic tools to analyze common error modes in SDS tasks. The proposed method is evaluated on the PASCAL VOC dataset, showing significant improvements over existing methods.The paper introduces a novel approach called Simultaneous Detection and Segmentation (SDS) to detect all instances of a category in an image and segment the pixels that belong to each instance. Unlike traditional bounding box detection or semantic segmentation, SDS requires both detection and precise pixel-level segmentation. The authors build on the Region Proposal Network (R-CNN) by using convolutional neural networks (CNNs) to classify category-independent region proposals and refine these proposals using category-specific figure-ground predictions. They achieve a 7-point boost in SDS performance, a 5-point boost in semantic segmentation, and state-of-the-art performance in object detection. The paper also provides diagnostic tools to analyze common error modes in SDS tasks. The proposed method is evaluated on the PASCAL VOC dataset, showing significant improvements over existing methods.