Winners of Challenge

Rank

Model

Participant

Affiliation

Attempt
Date

CIDEr

ROUGE-L

SPICE

Human Ratings

(T2 Dataset)

1

3ensembles-
cider-opt

tticbiu

TTIC+BIU

4/26/19

0.9945 (2)

0.2875 (1)

0.2019 (1)

72.38%

2

1x64-16x64-
16x256-T2T-v2

Gail-
Captions-2

Google AI Language

4/26/19

0.9480 (3)

0.2620 (4)

0.1919 (4)

67.08%

3

1x64-16x64-
16x256-T2T-v1

Gail-
Captions

Google AI Language

4/25/19

0.9324 (4)

0.2630 (3)

0.1934 (3)

67.04%

4

5ensembles-
cross-entropy

tticbiu-2

TTIC+BIU

4/27/19

1.0412 (1)

0.2779 (2)

0.1977 (2)

66.39%

5

Transformer-
Baseline-single

Conceptual-
Challenge

Organizers

3/15/19

0.7724 (5)

0.2443 (5)

0.1720 (5)

62.54%

6

Single

THU-
Caption-
explorer

Department of Computer Science
and Technology, Tsinghua University

4/25/19

0.6572 (6)

0.2386 (6)

0.1576 (6)



T2 Dataset

Download human ratings for the above results here (link). Please note that the identity of the submitted models are removed in the human ratings downloads for contestants' anonymity.