Computer Vision Representation Research Articles

Neural structure fields with application to crystal structure autoencoders

Representing crystal structures of materials to facilitate determining them via neural networks is crucial for enabling machine-learning applications involving crystal structure estimation. Among these applications, the inverse design of materials can contribute to explore materials with desired properties without relying on luck or serendipity. Here, we propose neural structure fields (NeSF) as an accurate and practical approach for representing crystal structures using neural networks. Inspired by the concepts of vector fields in physics and implicit neural representations in computer vision, the proposed NeSF considers a crystal structure as a continuous field rather than as a discrete set of atoms. Unlike existing grid-based discretized spatial representations, the NeSF overcomes the tradeoff between spatial resolution and computational complexity and can represent any crystal structure. We propose an autoencoder of crystal structures that can recover various crystal structures, such as those of perovskite structure materials and cuprate superconductors. Extensive quantitative results demonstrate the superior performance of the NeSF compared with the existing grid-based approach.

Point cloud up-sampling network with multi-level spatial local feature aggregation

Point clouds is one of popular 3D representations in computer vision and computer graphics. However, due to the sparseness and non-uniformity, raw point cloud from scanning devices cannot applied to down-stream geometry analyzing tasks directly. In this paper, we propose an end-to-end point cloud up-sampling network to reconstruct the dense yet uniform-distributed point clouds. Firstly, we utilize the spatial relationship of local regions and capture point-wise features progressively. We then propose a novel network to aggregate those features from different levels. Finally, we design an up-sampling module which consists of multi-branch convolution units to generate the dense point clouds. We conduct sufficient experiments on currently available public benchmarks. Experimental results show that proposed method has achieved 0.103 and 0.010 performance on Hausdorff distance and Chamfer Distance on VisionAir dataset, in comparison with the baseline towards uniformity, proximity-to-surface and mesh reconstruction.

CNN Architectures for Geometric Transformation-Invariant Feature Representation in Computer Vision: A Review

CNN Architectures for Geometric Transformation-Invariant Feature Representation in Computer Vision: A Review

Self-supervised Learning: Generative or Contrastive

Deep supervised learning has achieved great success in the last decade. However, its deficiencies of dependence on manual labels and vulnerability to attacks have driven people to explore a better solution. As an alternative, self-supervised learning attracts many researchers for its soaring performance on representation learning in the last several years. Self-supervised representation learning leverages input data itself as supervision and benefits almost all types of downstream tasks. In this survey, we take a look into new self-supervised learning methods for representation in computer vision, natural language processing, and graph learning. We comprehensively review the existing empirical methods and summarize them into three main categories according to their objectives: generative, contrastive, and generative-contrastive (adversarial). We further investigate related theoretical analysis work to provide deeper thoughts on how self-supervised learning works. Finally, we briefly discuss open problems and future directions for self-supervised learning. An outline slide for the survey is provided.

Learning CNN features from DE features for EEG-based emotion recognition

Recently, deep neural networks (DNNs) have shown the remarkable success of feature representations in computer vision, audio analysis, and natural language processing. Furthermore, DNNs have been used for electroencephalography (EEG) signal classification in recent studies on brain–computer interface. However, most works use one-dimensional EEG features to learn DNNs that ignores the local information within multichannel or multiple frequency bands in the EEG signals. In this paper, we propose a novel emotion recognition method using a convolutional neural network (CNN) while preventing the loss of local information. The proposed method consists of two parts. The first part generates topology-preserving differential entropy features while keeping the distance from the center electrode to other electrodes. The second part learns the proposed CNN to estimate three-class emotional states (positive, neutral, negative). We evaluate our work on SEED dataset, including 62-channel EEG signals recorded from 15 subjects. Our experimental results demonstrate that the proposed method achieved superior performance on SEED dataset with an average accuracy of 90.41% with the visualization of extracted features from the proposed CNN using t-SNE to show our representation outperforms the other representations based on standard features for EEG analysis. Besides, with the additional experiment on VIG dataset to estimate the vigilance of EEG dataset, we show the off-the-shelf availability of the proposed method.

Evolving weighting schemes for the Bag of Visual Words

The Bag of Visual Words (BoVW) is an established representation in computer vision. Taking inspiration from text mining, this representation has proved to be very effective in many domains. However, in most cases, standard term-weighting schemes are adopted (e.g., term-frequency or TF-IDF). It remains open the question of whether alternative weighting schemes could boost the performance of methods based on BoVW. More importantly, it is unknown whether it is possible to automatically learn and determine effective weighting schemes from scratch. This paper brings some light into both of these unknowns. On the one hand, we report an evaluation of the most common weighting schemes used in text mining, but rarely used in computer vision tasks. Besides, we propose an evolutionary algorithm capable of automatically learning weighting schemes for computer vision problems. We report empirical results of an extensive study in several computer vision problems. Results show the usefulness of the proposed method.

Part-Machine Group Formation with Ordinal-Ratio Level Data & Production Volume

This work presents an algorithm for design of cellular manufacturing system. It considers ordinal level data and ratio level data. We consider sequence of operation for every component as ordinal level data. The ratio level data includes operation time per unit, production volume & machine capacity. Adapted design activity model & image representation in computer vision is used to model ordinal-ratio level data. Orthographic view part grouping is used for part family formation. Machines are allocated on basis of their net utilization for a part family. Performance of algorithm is comparable & of good quality with existing methods in its class. To validate the result, a two way ANOVA technique is used, and on simulated example the F-ratio is enumerated. It is observed that production volume has significant influence on cell configuration and cell performance. It is concluded that the change in production volume mixes may need reconfiguration & capacity scaling according to recommendation of algorithm & ANOVA analysis. To facilitate industrial application, it can be implemented by using free software, Scilab and/or commercial softwares.

A Framework for Generic Object Recognition with Bayesian Networks

Since Biederman introduced to the computer vision community a theory of human image understanding called “recognition-by-components,” great interest in using it as a basis for generic object recognition has been spawned. Inspired by OPTICA, we propose a framework for generic object recognition with multiple Bayesian networks, where the object, primitive, prediction, and face nets are integrated with the more commonly used graph representation in computer vision to capture the causal, probabilistic relations among objects, primitives, aspects, faces, and contours. Based on the use of likelihood evidence, the communication mechanism among the nets is simple and efficient, and the four basic recognition behaviours are realized in a single framework. Each net is an autonomous agent, selectively responding to the data from the lower level in the context from its parent net, and dealing with the uncertainty and controlling the recognition tasks on its corresponding level. Our contributions in this article are the dynamic feedback control among recognition stages based on Bayesian networks, the attention mechanism using consistency-and discrimination-based value functions, and the unification of incremental grouping, partial matching, and multi-key indexing as an identical process under prediction for hypothesis generation. Our experiments have demonstrated that this new approach is more robust and efficient than the previous one.