Billion-scale semi-supervised learning for image classification

I. Zeki YalnizHervé JégouKan ChenManohar PaluriDhruv Mahajan

   Papers with code   Abstract  PDF

This paper presents a study of semi-supervised learning with large convolutional networks. We propose a pipeline, based on a teacher/student paradigm, that leverages a large collection of unlabelled images (up to 1 billion)... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
ResNeXt-101 32x8d
(288x288 Mean-Max Pool…)
85.1%
--
4
ResNeXt-101 32x16d
(288x288 Mean-Max Pool…)
84.3%
--
5
ResNeXt-101 32x4d
(288x288 Mean-Max Pool…)
84.0%
--
10
ResNeXt-50 32x4d
(288x288 Mean-Max Pool…)
83.1%
--
11
ResNeXt-101 32x16d
(288x288 Mean-Max Pool…)
82.6%
--
12
ResNeXt-101 32x8d
(288x288 Mean-Max Pool…)
82.5%
--
14
ResNeXt-101 32x4d
(288x288 Mean-Max Pool…)
82.0%
--
15
ResNet-50
(288x288 Mean-Max Pool…)
82.0%
--
16
ResNeXt-50 32x4d
(288x288 Mean-Max Pool…)
81.3%
--
17
ResNet-50
(288x288 Mean-Max Pool…)
80.3%
--