We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Single-Modal Entropy based Active Learning for Visual Question Answering

Abstract: Constructing a large-scale labeled dataset in the real world, especially for high-level tasks (eg, Visual Question Answering), can be expensive and time-consuming. In addition, with the ever-growing amounts of data and architecture complexity, Active Learning has become an important aspect of computer vision research. In this work, we address Active Learning in the multi-modal setting of Visual Question Answering (VQA). In light of the multi-modal inputs, image and question, we propose a novel method for effective sample acquisition through the use of ad hoc single-modal branches for each input to leverage its information. Our mutual information based sample acquisition strategy Single-Modal Entropic Measure (SMEM) in addition to our self-distillation technique enables the sample acquisitor to exploit all present modalities and find the most informative samples. Our novel idea is simple to implement, cost-efficient, and readily adaptable to other multi-modal tasks. We confirm our findings on various VQA datasets through state-of-the-art performance by comparing to existing Active Learning baselines.
Comments: Accepted to BMVC 2021
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as: arXiv:2110.10906 [cs.CV]
  (or arXiv:2110.10906v2 [cs.CV] for this version)

Submission history

From: Dong-Jin Kim [view email]
[v1] Thu, 21 Oct 2021 05:38:45 GMT (2236kb,D)
[v2] Thu, 18 Nov 2021 06:44:13 GMT (2236kb,D)

Link back to: arXiv, form interface, contact.