Repeatability Is Not Enough: Learning Affine Regions via Discriminability (2017-11-17T00:00:00.000000Z)

TL;DR

It is shown that maximizing geometric repeatability does not lead to local regions, a.k.a features, that are reliably matched and this necessitates descriptor-based learning, and a novel hard negative-constant loss function is proposed for learning of affine regions.

Abstract

A method for learning local affine-covariant regions is presented. We show that maximizing geometric repeatability does not lead to local regions, a.k.a features, that are reliably matched and this necessitates descriptor-based learning. We explore factors that influence such learning and registration: the loss function, descriptor type, geometric parametrization and the trade-off between matchability and geometric accuracy and propose a novel hard negative-constant loss function for learning of affine regions. The affine shape estimator – AffNet – trained with the hard negative-constant loss outperforms the state-of-the-art in bag-of-words image retrieval and wide baseline stereo. The proposed training process does not require precisely geometrically aligned patches. The source codes and trained weights are available at https://github.com/ducha-aiki/affnet.

Authors

Dmytro Mishkin

5 papers

Jiri Matas

24 papers

Filip Radenovic

6 papers

TL;DR

Abstract

Authors

References65 items

Benchmarking 6DOF Outdoor Visual Localization in Changing Conditions

Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking

Radially-Distorted Conjugate Translations

Fine-Tuning CNN Image Retrieval with No Human Annotation

Automatic differentiation in PyTorch

Learning Spread-Out Local Feature Descriptors

Systematic evaluation of convolution neural network advances on the Imagenet

Benchmarking 6DOF Urban Visual Localization in Changing Conditions

L2-Net: Deep Learning of Discriminative Patch Descriptor in Euclidean Space

Comparative Evaluation of Hand-Crafted and Learned Local Features

Learning Discriminative and Transformation Covariant Local Feature Detectors

Working hard to know your neighbor's margins: Local descriptor learning loss

HPatches: A Benchmark and Evaluation of Handcrafted and Learned Local Descriptors

Designing Deep Convolutional Neural Networks for Continuous Object Orientation Estimation

Large-Scale Image Retrieval with Attentive Deep Local Features

Quad-Networks: Unsupervised Learning to Rank for Interest Point Detection

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

End-to-End Learning of Deep Visual Representations for Image Retrieval

Structure-from-Motion Revisited

Universal Correspondence Network

From Dusk Till Dawn: Modeling in the Dark

Learning Covariant Feature Detectors

CNN Image Retrieval Learns from BoW: Unsupervised Fine-Tuning with Hard Examples

LIFT: Learned Invariant Feature Transform

Deep Residual Learning for Image Recognition

Learning to Assign Orientations to Feature Points

MatchNet: Unifying feature and metric learning for patch-based matching

From single image query to detailed 3D reconstruction

Spatial Transformer Networks

WxBS: Wide Baseline Stereo Generalizations

Learning to compare image patches via convolutional neural networks

MODS: Fast and robust method for two-view matching

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Adam: A Method for Stochastic Optimization

TILDE: A Temporally Invariant Learned DEtector

Visual query expansion with or without geometry: Refining local descriptors by feature aggregation

Location recognition over large time lags

Predicting Matchability

Learning Vocabularies over a Fine Quantization

Three things everyone should know to improve object retrieval

Image matching using local symmetry features

Edge foci interest points

ORB: An efficient alternative to SIFT or SURF

Rectified Linear Units Improve Restricted Boltzmann Machines

Improving Bag-of-Features for Large Scale Image Search

On the burstiness of visual elements

Efficient representation of local geometry for large scale object retrieval

Lost in quantization: Improving particular object retrieval in large scale image databases

Registration of Challenging Image Pairs: Initialization, Estimation, and Decision

Automatic Panoramic Image Stitching using Invariant Features

Object retrieval with large vocabularies and fast spatial matching

Keypoint Descriptors for Matching Across Multiple Image Modalities and Non-linear Intensity Variations

A Comparison of Affine Region Detectors

Distinctive Image Features from Scale-Invariant Keypoints

Scale & Affine Invariant Interest Point Detectors

Video Google: a text retrieval approach to object matching in videos

Reliable feature matching across widely separated views

An Iterative Image Registration Technique with an Application to Stereo Vision

Learning local feature descriptors with triplets and shallow convolutional neural networks

Dropout: a simple way to prevent neural networks from overfitting

Fixing the Locally Optimized RANSAC

Accurate Image Search Using the Contextual Dissimilarity Measure

Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration

Ieee Transactions on Pattern Analysis and Machine Intelligence 1 Discriminative Unsupervised Feature Learning with Exemplar Convolutional Neural Networks

Repeatability Is Not Enough: Learning Afﬁne Regions via Discriminability 17

Field of Study

Venue Information

Name

Type

URL

Alternate Names