Structure Is a Visual Class Invariant

Xiao, Bai; Song, Yi-Zhe; Balika, Anupriya; Hall, Peter M.

doi:10.1007/978-3-540-89689-0_37

Cited by 3 publications

(3 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Earlier work [3,50,51] suggests that structure is invariant across depictive styles, and therefore useful for cross-depiction detection. As described in Section 4, Fast R-CNN includes an ROI pooling layer, which carries out max-pooling over H × W uniformly spaced rectangular grid.…”

Section: The Importance Of Global Structurementioning

confidence: 99%

Detecting People in Artwork with CNNs

Westlake

Cai

Hall

2016

Lecture Notes in Computer Science

View full text Add to dashboard Cite

CNNs have massively improved performance in object detection in photographs. However research into object detection in artwork remains limited. We show state-of-the-art performance on a challenging dataset, People-Art, which contains people from photos, cartoons and 41 different artwork movements. We achieve this high performance by fine-tuning a CNN for this task, thus also demonstrating that training CNNs on photos results in overfitting for photos: only the first three or four layers transfer from photos to artwork. Although the CNN's performance is the highest yet, it remains less than 60\% AP, suggesting further work is needed for the cross-depiction problem. The final publication is available at Springer via http://dx.doi.org/10.1007/978-3-319-46604-0_57Comment: 14 pages, plus 3 pages of references; 7 figures in ECCV 2016 Workshop

show abstract

Section: The Importance Of Global Structurementioning

confidence: 99%

Detecting People in Artwork with CNNs

Westlake

Cai

Hall

2016

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…Chem. Acta 2017, 90 (3), [359][360][361][362][363][364][365][366][367][368] The Laplacian energy, Equation (4), found a fully different and absolutely unforeseen area of application: in pattern recognition and image analysis, [181][182][183][184][185][186] which, in addition, is attempted to be used in medical investigations of brain activity. [187,188] The title of the paper [184] is characteristic: "High-resolution satellite image classification and segmentation using Laplacian graph energy".…”

Section: Doi: 105562/cca3189mentioning

confidence: 99%

The Total π-Electron Energy Saga

Gutman¹,

Furtula²

2017

Croat. Chem. Acta

View full text Add to dashboard Cite

Abstract:The total π-electron energy, as calculated within the Hückel tight-binding molecular orbital approximation, is a quantum-theoretical characteristic of conjugated molecules that has been conceived as early as in the 1930s. In 1978, a minor modification of the definition of total π-electron energy was put forward, that made this quantity interesting and attractive to mathematical investigations. The concept of graph energy, introduced in 1978, became an extensively studied graph-theoretical topic, with many hundreds of published papers. A great variety of graph energies is being considered in the current mathematical-chemistry and mathematical literature. Recently, some unexpected applications of these graph energies were discovered, in biology, medicine, and image processing.We provide historic, bibliographic, and statistical data on the research on total π-electron energy and graph energies, and outline its present state of art. The goal of this survey is to provide, for the first time, an as-complete-as-possible list of various existing variants of graph energy, and thus help the readers to avoid getting lost in the jungle of references on this topic.

show abstract

“…As originally used, SSD matching employs a "star graph" to match; we allow a more general structure and (as mentioned above and detailed below) augment SSDs with geometry. Evidence for the utility of structure to our problem come from Bai et al [20], who build a depiction invariant classifier using structure as the only feature. However, the classifier yields broad classes, which motivates our richer description of parts.…”

Section: Introductionmentioning

confidence: 99%

Depiction Inviariant Object Matching

Balikai

Hall

2012

Procedings of the British Machine Vision Conference 2012

View full text Add to dashboard Cite

We are interested in matching objects in photographs, paintings, sketches and so on; after all, humans have a remarkable ability to recognise objects in images, no matter how they are depicted. We conduct experiments in matching, and conclude that the key to robustness lies in object description. The existing literature consists of numerous feature descriptors that rely heavily on photometric properties such as colour and illumination to describe objects. Although these methods achieve high rates of accuracy in applications such as detection and retrieval of photographs, they fail to generalise datasets consisting of mixed depictions. Here, we propose a more general approach for describing objects invariant to depictive style. We use structure at a global level, which is combined with simple non-photometric descriptors at a local level. There is no need for any prior learning. Our descriptor achieves results on par with existing state of the art, when applied to object matching on a standard dataset consisting of photographs alone and outperforms the state of the art when applied to depiction-invariant object matching.

show abstract

Structure Is a Visual Class Invariant

Cited by 3 publications

References 18 publications

Detecting People in Artwork with CNNs

Detecting People in Artwork with CNNs

The Total π-Electron Energy Saga

Depiction Inviariant Object Matching

Contact Info

Product

Resources

About