Self-Erasing Network for Integral Object Attention

05/09/201813/05/2019 Peng-TaoJiang

Visual results on a complex image with 4 semantic categories

Code and results on VOC images are available [here]

Per-category IoU scores are available:

VOC val set

category	IoU (VGG)	IoU (ResNet-101)
background	0.890	0.897
aeroplane	0.824	0.858
bicycle	0.324	0.437
bird	0.715	0.775
boat	0.634	0.617
bottle	0.627	0.681
bus	0.803	0.760
car	0.728	0.668
cat	0.799	0.832
chair	0.178	0.173
cow	0.692	0.790
dining table	0.194	0.100
dog	0.736	0.805
horse	0.718	0.811
motorbike	0.693	0.744
person	0.683	0.757
potted plant	0.355	0.389
sheep	0.730	0.774
sofa	0.266	0.196
train	0.735	0.676
tv/monitor	0.502	0.507
mean	0.611	0.631

If you think this work is helpful for you, please cite:

@inproceedings{hou2018selferasing,
  title={Self-Erasing Network for Integral Object Attention},
  author={Hou, Qibin and Jiang, Peng-Tao and Wei, Yunchao and Cheng, Ming-Ming},
  booktitle={NeurIPS},
  year={2018}
}

(Visited 4,096 times, 1 visits today)