FindAPhD Weekly PhD Newsletter | JOIN NOW FindAPhD Weekly PhD Newsletter | JOIN NOW

Graphs of Deep Networks


   Department of Computer Science

This project is no longer listed on FindAPhD.com and may not be available.

Click here to search FindAPhD.com for PhD studentship opportunities
  Dr A Bors  Applications accepted all year round  Self-Funded PhD Students Only

About the Project

This project aims to develop a new approach for joint syntactical and statistical data representation modelling. Image or video data representing a scene can be modelled by a graph [1] whose nodes are represented by Convolution Neural Networks (CNN), such as ResNet-50 [2], or they can be represented by relationships between Transformer networks [11], resulting in Graph Convolution Neural Networks (GCN) or Graphs of Transformer Networks (GTN). The scene representation can be performed following identification of objects or regions [3] or scene segmentation [4]. The relationships between these regions are modelled by the edges connecting the nodes, while their associated weights indicate the degree of connectiveness [5,6] or interactions, either static or through movement [7,8]. The inter-dependencies between the features extracted by each CNN network can be defined through adjacency tensors [5,8]. An orthogonal decomposition can then be used to extract a syntactical image representation [4,9]. Data representations can be defined hierarchically [2,4,7], where the significance in the scene is modelled by the upper nodes. Your approach will have to define how the networks in the GCN interact with each other for the common goal [9]. A generative tree model, GAN-Tree was proposed in [10]. An interesting approach would be to use Transformers such as the Swin Transformer [11] for defining GTNs.

This project will develop a graph representation for images/video, by defining criteria for assigning processing nodes to specific regions in scenes and optimizing the modelling of the interconnectivities between such regions.

Objectives: Define a graph convolution network (CGN), jointly syntactical and statistical image representations, cost function for training CGN, minimizing the total number of parameters and computational complexity required.

Applications: Scene representation and understanding, person & vehicle reidentification, retrieval, scene synthesis, matching occluded images

Research areas: Computer Vision and Image Processing; Machine learning; Neural networks;

Applications: Syntactic scene representation from images/video, Recognition and Classification, Content Based Image retrieval, Vehicle/Object re-identification.

The candidate should be familiar or willing to learn about deep learning tools such as PyTorch or TensorFlow.


References

[1] B. Jiang, X. Wang, and B. Luo, PH-GCN: Person Re-identification with Part-based Hierarchical Graph Convolutional Network, https://arxiv.org/abs/1907.08822, 2019.
[2] K. He, et al., Deep residual learning for image recognition, Proc. CVPR, 2016, pp. 770-778.
[3] K. He et al., Mask R-CNN. Proc. ICCV, 2017, pp. 2961-2969.
[4] X. Li et at., Spatial Pyramid Based Graph Reasoning for Semantic Segmentation, Proc. CVPR, 2020, pp. 8947–8956.
[5] Z. M. Chen et al., Multi-Label Image Recognition with Graph Convolutional Networks, Proc. CVPR, 2019, pp. 5172-5181.
[6] Y. Shen et al., Person Re-identification with Deep Similarity-Guided Graph Neural Network, Proc. ECCV, vol. LNCS-11219, 2018, pp 508-526.
[7] J. Zhang et al., Temporal Reasoning Graph for Activity Recognition, IEEE Trans. on Image Processing, vol. 29, no. 4, pp. 5491-5506, 2020.
[8] R. Herzig et al., Spatio-Temporal Action Graph Networks, Proc. ICCV-w, 2019, pp. 2347-2356.
[9] Z, Xu et. al.," HSS-GCN: A Hierarchical Spatial Structural Graph Convolutional Network for Vehicle Re-identification," Proc. ICPR-w (IUC), vol. LNCS vol. 12665, pp 356-364, 2021.
[10] J. N. Kundu et al. GAN-Tree: An Incrementally Learned Hierarchical Generative Framework for Multi-Modal Data Distributions, Proc. ICCV, 2019, pp. 8191-8200.
[11] Z. Li et al. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows, Proc. ICCV, 2021.

How good is research at University of York in Computer Science and Informatics?


Research output data provided by the Research Excellence Framework (REF)

Click here to see the results for all UK universities
Search Suggestions
Search suggestions

Based on your current searches we recommend the following search filters.

PhD saved successfully
View saved PhDs