Multiple instance learning (MIL) can reduce the need for costly annotation in
tasks such as semantic segmentation by weakening the required degree of
supervision. We propose a novel MIL formulation of multi-class semantic
segmentation learning by a fully convolutional network. In this setting, we
seek to learn a semantic segmentation model from just weak image-level labels.
The model is trained end-to-end to jointly optimize the representation while
disambiguating the pixel-image label assignment. Fully convolutional training
accepts inputs of any size, does not need object proposal pre-processing, and
offers a pixelwise loss map for selecting latent instances. Our multi-class MIL
loss exploits the further supervision given by images with multiple labels. We
evaluate this approach through preliminary experiments on the PASCAL VOC
segmentation challenge.
%0 Generic
%1 pathak2014fully
%A Pathak, Deepak
%A Shelhamer, Evan
%A Long, Jonathan
%A Darrell, Trevor
%D 2014
%K image label weakly
%T Fully Convolutional Multi-Class Multiple Instance Learning
%U http://arxiv.org/abs/1412.7144
%X Multiple instance learning (MIL) can reduce the need for costly annotation in
tasks such as semantic segmentation by weakening the required degree of
supervision. We propose a novel MIL formulation of multi-class semantic
segmentation learning by a fully convolutional network. In this setting, we
seek to learn a semantic segmentation model from just weak image-level labels.
The model is trained end-to-end to jointly optimize the representation while
disambiguating the pixel-image label assignment. Fully convolutional training
accepts inputs of any size, does not need object proposal pre-processing, and
offers a pixelwise loss map for selecting latent instances. Our multi-class MIL
loss exploits the further supervision given by images with multiple labels. We
evaluate this approach through preliminary experiments on the PASCAL VOC
segmentation challenge.
@misc{pathak2014fully,
abstract = {Multiple instance learning (MIL) can reduce the need for costly annotation in
tasks such as semantic segmentation by weakening the required degree of
supervision. We propose a novel MIL formulation of multi-class semantic
segmentation learning by a fully convolutional network. In this setting, we
seek to learn a semantic segmentation model from just weak image-level labels.
The model is trained end-to-end to jointly optimize the representation while
disambiguating the pixel-image label assignment. Fully convolutional training
accepts inputs of any size, does not need object proposal pre-processing, and
offers a pixelwise loss map for selecting latent instances. Our multi-class MIL
loss exploits the further supervision given by images with multiple labels. We
evaluate this approach through preliminary experiments on the PASCAL VOC
segmentation challenge.},
added-at = {2018-09-19T18:40:09.000+0200},
author = {Pathak, Deepak and Shelhamer, Evan and Long, Jonathan and Darrell, Trevor},
biburl = {https://www.bibsonomy.org/bibtex/28fea75ab178e98d285c040220586af12/huntstart},
description = {[1412.7144] Fully Convolutional Multi-Class Multiple Instance Learning},
interhash = {cec2b271d78548b6c1149724dce852b5},
intrahash = {8fea75ab178e98d285c040220586af12},
keywords = {image label weakly},
note = {cite arxiv:1412.7144Comment: in ICLR 2015},
timestamp = {2018-09-19T19:56:44.000+0200},
title = {Fully Convolutional Multi-Class Multiple Instance Learning},
url = {http://arxiv.org/abs/1412.7144},
year = 2014
}