Exploring the Limits of Weakly Supervised Pretraining

Dhruv MahajanRoss GirshickVignesh RamanathanKaiming HeManohar PaluriYixuan LiAshwin BharambeLaurens van der Maaten

   Papers with code   Abstract  PDF

State-of-the-art visual perception models for a wide range of tasks rely on supervised pretraining. ImageNet classification is the de facto pretraining task for these models... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
ResNeXt-101 32x48d
(288x288 Mean-Max Pool…)
86.1%
--
2
ResNeXt-101 32x32d
(288x288 Mean-Max Pool…)
85.9%
--
3
ResNeXt-101 32x48d
85.4%
85.4%
4
ResNeXt-101 32x48d
85.4%
85.4%
5
ResNeXt-101 32x32d
85.1%
85.1%
6
ResNeXt-101 32x32d
85.1%
85.1%
7
ResNeXt-101 32x16d
(288x288 Mean-Max Pool…)
85.0%
--
9
ResNeXt-101 32x16d
84.2%
--
10
ResNeXt-101 32x8d
(288x288 Mean-Max Pool…)
83.5%
--
11
ResNeXt-101 32x8d
82.7%
82.2%
12
ResNeXt-101 32x8d
82.7%
82.2%