Weakly Supervised Instance Segmentation by Deep Community Learning

Abstract

We present a weakly supervised instance segmentation algorithm based on deep community learning with multiple tasks. This task is formulated as a combination of weakly supervised object detection and semantic segmentation, where individual objects of the same class are identified and segmented separately. We address this problem by designing a unified deep neural network architecture, which has a positive feedback loop of object detection with bounding box regression, instance mask generation, instance segmentation, and feature extraction. Each component of the network makes active interactions with others to improve accuracy, and the end-to-end trainability of our model makes our results more robust and reproducible. The proposed algorithm achieves state-of-the-art performance in the weakly supervised setting without any additional training such as Fast R-CNN and Mask R-CNN on the standard benchmark dataset.

Network

Instance Segmentation Results on PASCAL VOC 2012

Method	Supervision	Post-processing	mAP0.25	mAP0.5	mAP0.75	ABO
WISE w/ Mask R-CNN	I	✓	49.2	41.7	23.7	55.2
IRN w/ Mask R-CNN	I	✓	–	46.7	–	–
Cholakkal et al.	I + C	✓	48.5	30.2	14.4	44.3
PRM	I	✓	44.3	26.8	9.0	37.6
IAM	I	✓	45.9	28.3	11.9	41.9
Label-PEnet	I	✓	49.2	30.2	12.9	41.4
Ours	I		57.0	35.9	5.8	43.8
Ours	I	✓	56.6	38.1	12.3	48.2

Qualitative Results

Video

Citation

@inproceedings{hwang2021weakly,
    author    = {Hwang, Jaedong and Kim, Seohyun and Son, Jeany and Han, Bohyung},
    title     = {Weakly Supervised Instance Segmentation by Deep Community Learning},
    booktitle = {WACV},
    year      = {2021},
}