Filters








221 Hits in 8.2 sec

Efficient Diffusion on Region Manifolds: Recovering Small Objects with Compact CNN Representations [article]

Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, Teddy Furon, Ondrej Chum
2019 arXiv   pre-print
Experimentally, we observe a significant boost in performance of image retrieval with compact CNN descriptors on standard benchmarks, especially when the query object covers only a small part of the image  ...  Small objects have been a common failure case of CNN-based retrieval.  ...  AQE is also effective with CNN global representation [52, 30, 18] . A baseline for the regional scenario is R-match [44] .  ... 
arXiv:1611.05113v3 fatcat:cxyn3wszn5gltmqlmvrzfnfjhq

Efficient Diffusion on Region Manifolds: Recovering Small Objects with Compact CNN Representations

Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, Teddy Furon, Ondrej Chum
2017 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
Experimentally, we observe a significant boost in performance of image retrieval with compact CNN descriptors on standard benchmarks, especially when the query object covers only a small part of the image  ...  Small objects have been a common failure case of CNN-based retrieval.  ...  AQE is also effective with CNN global representation [52, 30, 18] . A baseline for the regional scenario is R-match [44] .  ... 
doi:10.1109/cvpr.2017.105 dblp:conf/cvpr/IscenTAFC17 fatcat:3n246p5dnzeujf44cknb5xegki

Computational Analysis of Deformable Manifolds: from Geometric Modelling to Deep Learning [article]

Stefan C Schonsheck
2020 arXiv   pre-print
Next, we will use ideas from parallel transport on manifolds to generalize convolution and convolutional neural networks to deformable manifolds.  ...  More specifically, we will study techniques for representing manifolds and signals supported on them through a variety of mathematical tools including, but not limited to, computational differential geometry  ...  Similarly, a compact manifold is one which can be covered with a countable number of charts, following from the usual definition of a compact topological space being one for which every open cover has  ... 
arXiv:2009.01786v1 fatcat:ohqjzwldqnhafmtsb57f7otfzu

Implicit-PDF: Non-Parametric Representation of Probability Distributions on the Rotation Manifold [article]

Kieran Murphy, Carlos Esteves, Varun Jampani, Srikumar Ramalingam, Ameesh Makadia
2022 arXiv   pre-print
This is the most general way of representing distributions on manifolds, and to showcase the rich expressive power, we introduce a dataset of challenging symmetric and nearly-symmetric objects.  ...  We require no supervision on pose uncertainty -- the model trains only with a single pose per example.  ...  We go one step further and break the symmetry of objects by texturing with small markers (SYMSOL II).  ... 
arXiv:2106.05965v2 fatcat:jphaxm5smjc6ddbc4kqvi2jgwe

Diagnosing and Fixing Manifold Overfitting in Deep Generative Models [article]

Gabriel Loaiza-Ganem, Brendan Leigh Ross, Jesse C. Cresswell, Anthony L. Caterini
2022 arXiv   pre-print
This formulation directly contradicts the manifold hypothesis, which states that observed data lies on a low-dimensional manifold embedded in high-dimensional ambient space.  ...  We formally prove that degenerate optima are achieved wherein the manifold itself is learned but not the distribution on it, a phenomenon we call manifold overfitting.  ...  We weight the adversarial loss with a coefficient of 10.  ... 
arXiv:2204.07172v3 fatcat:4c74ka3g7jgohnyuqwltrxdvhy

Local Features and Visual Words Emerge in Activations [article]

Oriane Siméoni, Yannis Avrithis, Ondrej Chum
2019 arXiv   pre-print
The highest gain in performance is achieved when diffusion on the nearest-neighbor graph of global descriptors is initiated from spatially verified images.  ...  Initial ranking is based on image descriptors extracted from convolutional neural network activations by global pooling, as in recent state-of-the-art work.  ...  The other relies on a single global or few regional descriptors per image, leading to compact storage, efficient nearest neighbor search, and graph-based re-ranking.  ... 
arXiv:1905.06358v1 fatcat:77vxuxpmazh3ddjipg3brtoiei

Local Features and Visual Words Emerge in Activations

Oriane Simeoni, Yannis Avrithis, Ondrej Chum
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
The highest gain in performance is achieved when diffusion on the nearest-neighbor graph of global descriptors is initiated from spatially verified images.  ...  Initial ranking is based on image descriptors extracted from convolutional neural network activations by global pooling, as in recent state-ofthe-art work.  ...  The other relies on a single global or few regional descriptors per image, leading to compact storage, efficient nearest neighbor search, and graph-based re-ranking.  ... 
doi:10.1109/cvpr.2019.01192 dblp:conf/cvpr/SimeoniAC19 fatcat:r2kl56tjzfhtfnisqx2f37dxna

MeshCNN: A Network with an Edge [article]

Rana Hanocka, Amir Hertz, Noa Fish, Raja Giryes, Shachar Fleishman and Daniel Cohen-Or
2019 arXiv   pre-print
Polygonal meshes provide an efficient representation for 3D shapes.  ...  Analogous to classic CNNs, MeshCNN combines specialized convolution and pooling layers that operate on the mesh edges, by leveraging their intrinsic geodesic connections.  ...  Right: adaptive non-uniform representation. Large flat regions can be represented by a small number of large polygons, with detailed regions represented by a larger number of small polygons.  ... 
arXiv:1809.05910v2 fatcat:di5f53ex25hp3dgd7l7je4nlt4

Weakly Supervised Fine-Grained Image Classification via Guassian Mixture Model Oriented Discriminative Learning

Zhihui Wang, Shijie Wang, Shuhui Yang, Haojie Li, Jianjun Li, Zezhou Li
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
regions can be located more precisely on the new low-rank feature maps.  ...  By recovering the low-rank discriminative bases into the same embedding space of highlevel feature maps, LR 2 M alleviates the discriminative region diffusion problem in high-level feature map and discriminative  ...  Table 6 . 6 Comparison with the efficiency and effectiveness of other method on CUB-200-2011. K means the number of selected discriminative regions for each image.  ... 
doi:10.1109/cvpr42600.2020.00977 dblp:conf/cvpr/WangWYLLL20 fatcat:decxlubel5axhmpbnglj4jkexi

A Survey on Deep Visual Place Recognition

Carlo Masone, Barbara Caputo
2021 IEEE Access  
This was the first method that demonstrated efficient image retrieval, although on a small sized database.  ...  A major step towards overcoming these issues was the Regional Diffusion algorithm by Iscen et al. [140] , which included several strategies to make the diffusion refinement more efficient.  ... 
doi:10.1109/access.2021.3054937 fatcat:hc5fp2z4g5fldl7imkt7ixq4z4

3D Photography Using Context-Aware Layered Depth Inpainting

Meng-Li Shih, Shih-Yang Su, Johannes Kopf, Jia-Bin Huang
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Color and depth inpainting using diffusion is better, but provides a too smooth appearance (c).  ...  In contrast, our focus is on recovering the depth of the hidden surface. CNN-based single depth estimation.  ...  Our representation, on the other hand, can be easily converted into a textured mesh and efficiently rendered with standard graphics engines. Image inpainting.  ... 
doi:10.1109/cvpr42600.2020.00805 dblp:conf/cvpr/ShihSKH20 fatcat:u75konk4xnasfcyprhpxwigtva

A Literature Review: Geometric Methods and Their Applications in Human-Related Analysis

Wenjuan Gong, Bin Zhang, Chaoqi Wang, Hanbing Yue, Chuantao Li, Linjie Xing, Yu Qiao, Weishan Zhang, Faming Gong
2019 Sensors  
This review proposes to categorize geometric methods based on the scope of the geometric properties that are extracted: object-oriented geometric methods, feature-oriented geometric methods, and routine-based  ...  Geometric features, such as the topological and manifold properties, are utilized to extract geometric properties.  ...  Heat Diffusion CNN for Human Shape Analysis The heat diffusion equation is also used for extending traditional CNN to a manifold. Heat diffusion measures heat diffused on a manifold.  ... 
doi:10.3390/s19122809 fatcat:nrnq4vrj7bfv7iv4aem5g6k4zu

From Faces to Outdoor Light Probes

Dan A. Calian, Jean-François Lalonde, Paulo Gotardo, Tomas Simon, Iain Matthews, Kenny Mitchell
2018 Computer graphics forum (Print)  
We build compact, realistic representations of outdoor lighting both parametrically and in a data-driven way, by training a deep convolutional autoencoder on a large dataset of HDR sky environment maps  ...  We show that relighting objects with HDR light probes estimated by our method yields realistic results in a wide variety of settings.  ...  In [RRF * 15], "orientation-dependent appearance" of an object is estimated from a single image using a CNN with an architecture similar to ours.  ... 
doi:10.1111/cgf.13341 fatcat:vstx4sp5svcc3lnsn5hlb5hxk4

Subject-adaptive Integration of Multiple SICE Brain Networks with Different Sparsity

Jianjia Zhang, Luping Zhou, Lei Wang
2017 Pattern Recognition  
While enjoying beneficial properties, covariance representations also bring challenges. Both covariance descriptor and SICE matrix belong to the set of symmetric positive-definite (SPD) ma-  ...  In many contexts of computer vision, the data are represented by or converted to covariance-based representations, including covariance descriptor and sparse inverse covariance estimation (SICE), due to  ...  Result on object classification We further investigate the effectiveness of the proposed kernel representation on the tasks traditionally applied with covariance matrix as a region descriptor.  ... 
doi:10.1016/j.patcog.2016.09.024 fatcat:mkwvr4jtxbgyjb2fxjcszp3d4i

PDE-based Group Equivariant Convolutional Neural Networks [article]

Bart Smets, Jim Portegies, Erik Bekkers, Remco Duits
2022 arXiv   pre-print
Formulating our PDEs on homogeneous spaces allows these networks to be designed with built-in symmetries such as rotation in addition to the standard translation equivariance of CNNs.  ...  CNNs.  ...  Although steerable operators have clear benefits in terms of computational efficiency and accuracy [59, 60] , working with steerable representations puts constraints on non-linear activations within the  ... 
arXiv:2001.09046v6 fatcat:xywajlp75fadxmttoj5gayyhpy
« Previous Showing results 1 — 15 out of 221 results