Indexed Operations for Non-rectangular Lattices Applied to Convolutional Neural Networks

Jacquemont, Mikaël; Antiga, Luca; Vuillaume, Thomas; Silvestri, Giorgia; Benoît, Alexandre; Lambert, Patrick; Maurin, G.

doi:10.5220/0007364303620371

Cited by 16 publications

(12 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The baseline models we choose include normal VGG-VD-16 [15], binary patterns encoded CNN [66], two-stream deep fusion CNN [67], and hexagon based HEXACONV [10], IndexedConv [51]. Our experiments are conducted with 80%/20% splits to obtain better estimates and fair comparison.…”

Section: B: Results and Analysismentioning

confidence: 99%

See 1 more Smart Citation

Hexagonal Convolutional Neural Networks for Hexagonal Grids

Luo

Zhang

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Hexagonal grids use a hierarchical subdivision tessellation to cover the entire plane or sphere. Due to the 6-fold rotational symmetry, hexagonal grids have some advantages (e.g. isoperimetry, equidistant neighbors, and uniform connectivity) over quadrangular and triangular girds, which makes them suitable to tackle tasks of geospatial information processing and intelligent decision-making. In this paper, we first introduce some applications based on the hexagonal grids. Then, we introduce the planer and spherical hexagonal grids and analyze the group representations for them, we review geometric deep learning, some Convolutional Neural Networks (CNNs) for hexagonal grids, and group-based equivariant convolution. Next in importance, we propose the HexagonNet for hexagonal grids, and define a new convolution operator and pooling operator. Finally, in order to evaluate the effectiveness of the proposed HexagonNet, we perform experiments on two tasks: aerial scene classification on the aerial image dataset (AID), and 3D shape classification on the ModelNet40 dataset. The experimental results verify the practical applicability of the HexagonNet given some fixed parameter budgets.

show abstract

Section: B: Results and Analysismentioning

confidence: 99%

“…H-CNN devises a framework which combines the local prediction with hexagon-based ensemble mechanism [9]. IndexedConv [51] presents indexed convolution and pooling for the indexed hexagonal grids.…”

Section: Geometric Deep Learningmentioning

confidence: 99%

Hexagonal Convolutional Neural Networks for Hexagonal Grids

Luo

Zhang

et al. 2019

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Even though it is not required for higher-level analysis, physics shows that this parameter provides meaningful information to solve the other tasks. The baseline backbone of γ-PhysNet is the convolutional part of a ResNet-56 [7,8], CIFAR-10 version, with full pre-activation implemented with IndexedConv [11]. The latter allows the direct processing of the hexagonal pixel images of the LST1 data used in this paper without transforming them to square pixel ones.…”

Section: γ-Physnet Architecturementioning

confidence: 99%

Deep Learning for Astrophysics, Understanding the Impact of Attention on Variability Induced by Parameter Initialization

Jacquemont

Vuillaume

Benoît

et al. 2021

Pattern Recognition. ICPR International Workshops and Challenges

Self Cite

View full text Add to dashboard Cite

In the astrophysics domain, the detection and description of gamma rays is a research direction for our understanding of the universe. Gamma-ray reconstruction from Cherenkov telescope data is multi-task by nature. The image recorded in the Cherenkov camera pixels relates to the type, energy, incoming direction and distance of a particle from a telescope observation. We propose γ-PhysNet, a physically inspired multi-task deep neural network for gamma/proton particle classification, and gamma energy and direction reconstruction. As ground truth does not exist for real data, γ-PhysNet is trained and evaluated on large-scale Monte Carlo simulations. Robustness is then crucial for the transfer of the performance to real data. Relying on a visual explanation method, we evaluate the influence of attention on the variability due to weight initialization, and how it helps improve the robustness of the model. All the experiments are conducted in the context of single telescope analysis for the Cherenkov Telescope Array simulated data analysis.

show abstract

“…It is a multi-task network, so contrary to the classical approach and to previous works using CNNs, a single algorithm is trained to perform the full event reconstruction (particle classification, energy reconstruction and direction reconstruction). It is composed of a first convolution block common to all tasks and based on ResNet-56 [18] enhanced with indexed convolutions [19] and Dual Attention mechanism [20].…”

Section: Ii-b -Physnet -A Convolutional Neural Network To Analyse Lst-1 Datamentioning

confidence: 99%

Analysis of the Cherenkov Telescope Array first Large Size Telescope real data using convolutional neural networks

Abe¹,

Aguasca²,

Agudo³

et al. 2021

Proceedings of 37th International Cosmic Ray Conference — PoS(ICRC2021)

Self Cite

View full text Add to dashboard Cite

Indexed Operations for Non-rectangular Lattices Applied to Convolutional Neural Networks

Cited by 16 publications

References 0 publications

Hexagonal Convolutional Neural Networks for Hexagonal Grids

Hexagonal Convolutional Neural Networks for Hexagonal Grids

Deep Learning for Astrophysics, Understanding the Impact of Attention on Variability Induced by Parameter Initialization

Analysis of the Cherenkov Telescope Array first Large Size Telescope real data using convolutional neural networks

Contact Info

Product

Resources

About