Call For Papers

International Journal of Computer Vision

Special Issue on Efficient Visual Recognition



Visual recognition plays a central role in computer vision. A large amount of vision tasks fundamentally rely on the ability to recognize and localize faces, people, objects, scenes, places, attributes, actions and relations. Visual recognition thus touches many areas of artificial intelligence and information retrieval, such as image search, visual surveillance, video data mining, question answering, autonomous driving and robotic interactions.


Feature representation is the core of visual recognition. Milestone handcrafted feature descriptors such as Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP) have dominated visual recognition for years until the turning point in 2012 when Deep Convolutional Neural Networks (DCNN) achieved a record-breaking image classification accuracy. Since DCNN entered the scene, visual recognition has been experiencing a revolution and tremendous progress (such as enabling superhuman accuracy) has been achieved because of the availability of large visual datasets and GPU computing resources. Hand in hand went, the development of deeper and larger DCNNs that could automatically learn more and more powerful feature representations with multiple levels of abstraction from big data.


In many real world applications, recognizing efficiently is as critical as recognizing accurately. Significant progress has been made in the past few years to boost the accuracy levels of visual recognition, but existing solutions often rely on computationally expensive feature representation and learning approaches, which are too slow for numerous applications. In addition to the opportunities they offer, the large visual datasets also lead to the challenge of scalling up while retaining the efficiency of learning approaches and representations for both hand crafted and deeply learned features.


In addition, given sufficient amount of annotated visual data, some existing features, especially DCNN features, have been shown to yield high accuracy for visual recognition. However, there are many applications where only limited amounts of annotated training data can be available or collecting labeled training data is too expensive. Such applications impose great challenges to many existing features. Finally, with the prevalence of social media networks and mobile/wearable devices which have limited computational capabilities and storage space, the demand for sophisticated mobile/wearable device applications in handling visual big data recognition are rising. In such applications, real time performance is of utmost importance to users, since no one is willing to spend time waiting nowadays. Therefore, there is a growing need for developing visual features that are fast to compute, memory efficient, and yet exhibiting good discriminability and robustness for visual recognition.




We encourage researchers to study and develop novel efficient visual recognition approaches that are computationally efficient, memory efficient, and yet exhibiting good recognition accuracy. We aim to solicit original contributions that: (1) present state of the art theories related to efficient visual recognition; (2) explore novel algorithms and applications; (3) survey the recent progress in this field; and (4) establish benchmark datasets.


The list of possible topics includes, but is not limited to:

§ Hashing/binary coding and its related applications

§ Compact and efficient convolutional neural networks

§ Efficient handcrafted feature design

§ Fast features tailored to wearable/mobile devices

§ Efficient dimensionality reduction and feature selection

§ Sparse representation and its related applications

§ Evaluations of current handcrafted descriptors and deep learning based features

§ DCNN compression/quantization/binarization

§ Hybrid methods combining strengths of handcrafted and learning based approaches

§ Efficient feature learning for applications with limited amounts of annotated training data

§ Efficient approaches to increase the invariance of DCNN

Priority will be given to papers with high novelty and originality for research papers, and to papers with high potential impact for survey/overview papers.


Paper Submission and Review:


Authors are encouraged to submit original work that has not appeared in, nor is in consideration by, other journals. Papers extending previously published conference papers can be submitted, as long as the journal submission provides a significant contribution beyond the conference paper (The overlap is described clearly at the beginning of the journal submission).


Manuscripts will be subject to a peer reviewing process and must conform to the author guidelines available on the IJCV website at Instructions for Authors on the right panel.


Authors need to submit full papers online through the IJCV submission site at,, selecting the choice that indicates this special issue Efficient Visual Recognition.


Time line:


Submission of full papers: February 15, 2019.


Guest editors


·       Li Liu, National University of Defense Technology, China,

University of Oulu, Finland

·       Matti Pietikäinen, University of Oulu, Finland,

·       Jie Qin, ETH Zürich, Switzerland,

·       Jie Chen, University of Oulu, Finland,

·       Wanli Ouyang, University of Sydney, Australia,

·       Luc Van Gool,  ETH Zürich, Switzerland,




Li Liu


National University of Defense Technology, China

Center for Machine Vision and Signal Analysis (CMVS),
University of Oulu, Finland