1. Introduction
  2. Related Work
  3. Our approach
    1. Proposal generation
    2. Feature extraction
    3. Region classification
    4. Region refinement
  4. Experiments and results
    1. Results on \(AP^r\) and \(AP^r_{vol}\)
    2. Producing diagnostic information
    3. Results on \(AP^b\) and \(AP^b_{vol}\)
    4. Results on pixel IU

(CVPR 2016) Multi-scale Patch Aggregation (MPA) for Simultaneous Detection and Segmentation
Paper: http://www.cse.cuhk.edu.hk/leojia/papers/mpa_cvpr16.pdf

Introduction

Contributions:

  • generate dense multi-scale patches for object parsing.

  • unified end-to-end trainable proposal-free network can achieve segmentation and classification simultaneously for each patch.

  • develop an efficient algorithm to infer the segmentation mask for each object by merging information from mid-level patches.

Our approach

Proposal generation

Feature extraction

Region classification

Region refinement

Experiments and results

Results on \(AP^r\) and \(AP^r_{vol}\)

Producing diagnostic information

Results on \(AP^b\) and \(AP^b_{vol}\)

Results on pixel IU