BMask RCNN

Motivation:

Objectives:

Related Works:

• Existing methods can be divided into two categories:

  1. Detection-based methods: employ object detectors [15,18,34,44] to generate region proposals and then predict their masks after RoI pooling/align [18,21]. Based on CNN, [13,42,43] predict masks for object proposals. FCIS [33] extends InstanceFCN [13] by exploiting position-sensitive inside/outside score maps and fully convolutional networks for instance segmentation. BAIS [20] uses boundary-based distance transform to predict mask pixels that are beyond bounding boxes. Mask R-CNN [21] extends Faster R-CNN [44] by adding a mask prediction branch in parallel with the existing box regression and classification branches, demonstrating competitive performance on both object detection and instance segmentation. PANet [37] based on Mask R-CNN introduces the bottom-up path augmentation for FPN [34] to enhance information flow and adaptive feature pooling for better mask features. Mask scoring R-CNN [24] addresses the misalignment between mask quality and mask score in Mask R-CNN by explicitly learning the quality of predicted masks. [7] further improves Cascade Mask R-CNN [5] by interweaving box and mask branches in a multi-stage cascade manner and providing spatial context through semantic segmentation. Huang et al. apply a criss-cross attention module [25] to capture the full-image contextual information for instance segmentation. [30] draws on the idea of rendering and adaptively selects key points to recover fine details for high-quality image segmentation. 

  2. Segmentation-based methods: first exploit pixel-level segmentation over the image and then group the pixels together for each object. InstanceCut [29] adopts boundaries to partition semantic segmentation into instance-level segmentation. SGN [36] groups pixels along rows and columns by line segments. [47] utilizes predicted instance centers and pixel-wise directions to group instances. Recently, several methods [4,17] take the advantage of deep metric learning to learnt the embedding to group pixels to for instance segmentation.

Contribution:

Method 1:

Method 2:

Dataset:

Metrics:

Experimental Results:

Ablations:

References:

1. • n2  n0

   • θ