Open-ended Visual Question Answering on VQA (test-standard)

83.3Accuracy (Overall)

Human

Updated 1mo ago

Evaluation Results

Method
Human 2016.06	83.3	95.8	83.4	72.7
Ensemble of 7 Att. models 2016.06	66.5	83.2	39.5	58
Naver Labs (challenge 2nd) 2016.06	64.8	83.3	38.7	54.6
Ours-ResNet 2016.06	63.2	81.7	38.2	52.8
HieCoAtt 2016.06	62.1	-	-	-
(Lu et al. 2016) ResNet* 2016.06	62.1	-	-	-
Hierarchical Co-Attention (Alternating) + ResNet 2016.05	62.1	-	-	-
(Kim et al. 2016) ResNet* 2016.06	61.8	82.4	38.2	49.4
DMN+ 2016.06	60.4	-	-	-
(Xiong, Merity, and Socher 2016) 2016.06	60.4	80.4	36.8	48.3
DMN+ 2016.05	60.4	-	-	-
FDA 2016.06	59.5	-	-	-
FDA 2016.05	59.5	-	-	-
D-NMN 2016.06	59.4	-	-	-
AMA 2016.06	59.4	81.1	37.1	45.8
(Wu et al. 2016) 2016.06	59.4	81.1	37.1	45.8
SAN 2016.06	58.9	-	-	-
(Yang et al. 2016) 2016.06	58.9	-	-	-
SAN 2016.05	58.9	-	-	-
NMN 2016.06	58.7	81.2	37.7	44
AYN 2016.06	58.4	78.2	36.3	46.3
SMem 2016.06	58.2	80.9	37.5	43.5
VQA team 2016.06	58.2	80.6	36.5	43.7
(Antol et al. 2015) 2016.06	58.2	80.6	36.5	43.7
LSTM Q+I 2016.05	58.2	-	-	-
SMem 2016.05	58.2	-	-	-
(Andreas et al. 2016b) 2016.06	58	-	-	-
DPPnet 2016.06	57.4	80.3	36.9	42.2
(Noh, Seo, and Han 2016) 2016.06	57.4	80.3	36.9	42.2
iBOWIMG 2016.06	55.9	76.8	35	42.6
(Zhou et al. 2015) 2016.06	55.9	76.8	35	42.6
(Andreas et al. 2016a) 2016.06	55.1	-	-	-