Abstract | ||
---|---|---|
•We show that low-level semantic information can be derived from low-level CNN layers to improve VQA performance.•HFnet, a novel VQA that exploits both low and high-level features from a CNN for encoding visual semantics, is proposed.•The proposed method improves VQA accuracy using only the image-question paired dataset. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1016/j.neucom.2019.03.035 | Neurocomputing |
Keywords | Field | DocType |
Visual question answering,Multi-level features,Neural networks | Question answering,Convolutional neural network,Exploit,Artificial intelligence,Semantics,Machine learning,Mathematics | Journal |
Volume | ISSN | Citations |
351 | 0925-2312 | 1 |
PageRank | References | Authors |
0.40 | 0 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jongkwang Hong | 1 | 1 | 1.41 |
Jianlong Fu | 2 | 195 | 22.47 |
Youngjung Uh | 3 | 1 | 1.07 |
Tao Mei | 4 | 4702 | 288.54 |
Hyeran Byun | 5 | 505 | 65.97 |