Geo-Aware Networks for Fine-Grained Recognition

Chu, Grace; Potetz, Brian; Song, Yang; Brücher, Fernando; Leung, Thomas; Adam, Hartwig

doi:10.48550/arxiv.1906.01737

Cited by 1 publication

(9 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Baselines Our baselines include 1) direct directly applying feed-forward nets (Chu et al, 2019); 2) tile discretization (Berg et al, 2014;Adams et al, 2015;Tang et al, 2015); 3) wrap feed-forward nets with coordinate wrapping (Mac Aodha et al, 2019); and 4) rbf Radial Basis Function (RBF) kernels (Baudat & Anouar, 2001;Bierens, 1994). See Appendix A.1 for details of the baselines.…”

Section: Methodsmentioning

confidence: 99%

“…Recently the computer vision community shows increasing interests in incorporating geographic information (e.g. coordinate encoding) into neural network architectures for multiple tasks such as image classification (Tang et al, 2015) and fine grained recognition (Berg et al, 2014;Chu et al, 2019;Mac Aodha et al, 2019). Both Berg et al (2014) and Tang et al (2015) proposed to discretize the study area into regular grids.…”

Section: Related Workmentioning

confidence: 99%

“…In addition, discretization does not scale well in terms of memory use. To overcome these difficulties, both Chu et al (2019) and Mac Aodha et al ( 2019) advocated the idea of inductive location encoders which directly encode coordinates into a location embedding. However, both of them directly feed the coordinates into a feed-forward neural network (Chu et al, 2019) or residual blocks (Mac Aodha et al, 2019) without any feature decomposition strategy.…”

Section: Related Workmentioning

confidence: 99%

“…To overcome these difficulties, both Chu et al (2019) and Mac Aodha et al ( 2019) advocated the idea of inductive location encoders which directly encode coordinates into a location embedding. However, both of them directly feed the coordinates into a feed-forward neural network (Chu et al, 2019) or residual blocks (Mac Aodha et al, 2019) without any feature decomposition strategy. Our experiments show that this direct encoding approach is insufficient to capture the spatial feature distribution and Space2Vec significantly outperforms them by integrating spatial representations of different scales.…”

Section: Related Workmentioning

confidence: 99%

“…Li et al (2017) propose DCRNN for traffic forecasting in which the traffic sensor network is converted to a distance weighted graph which necessarily forfeits information about the spatial layout of sensors. There is, however, no general representation model beyond simply applying discretization (Berg et al, 2014;Tang et al, 2015) or feed-forward nets (Chu et al, 2019;Mac Aodha et al, 2019) to coordinates.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells

Mai,

Janowicz,

Yan

et al. 2020

Preprint

View full text Add to dashboard Cite

Unsupervised text encoding models have recently fueled substantial progress in Natural Language Processing (NLP). The key idea is to use neural networks to convert words in texts to vector space representations (embeddings) based on word positions in a sentence and their contexts, which are suitable for end-to-end training of downstream tasks. We see a strikingly similar situation in spatial analysis, which focuses on incorporating both absolute positions and spatial contexts of geographic objects such as Points of Interest (POIs) into models. A general-purpose representation model for space is valuable for a multitude of tasks. However, no such general model exists to date beyond simply applying discretization or feedforward nets to coordinates, and little effort has been put into jointly modeling distributions with vastly different characteristics, which commonly emerges from GIS data. Meanwhile, Nobel Prize-winning Neuroscience research shows that grid cells in mammals provide a multi-scale periodic representation that functions as a metric for location encoding and is critical for recognizing places and for path-integration. Therefore, we propose a representation learning model called Space2Vec to encode the absolute positions and spatial relationships of places. We conduct experiments on two real-world geographic data for two different tasks: 1) predicting types of POIs given their positions and context, 2) image classification leveraging their geo-locations. Results show that because of its multiscale representations, Space2Vec outperforms well-established ML approaches such as RBF kernels, multi-layer feed-forward nets, and tile embedding approaches for location modeling and image classification tasks. Detailed analysis shows that all baselines can at most well handle distribution at one scale but show poor performances in other scales. In contrast, Space2Vec 's multi-scale representation can handle distributions at different scales. 1

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells

Mai,

Janowicz,

Yan

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

Geo-Aware Networks for Fine-Grained Recognition

Cited by 1 publication

References 17 publications

Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells

Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells

Contact Info

Product

Resources

About