CBIR in academia and industry

Awesome image retrieval papers

The main goal is collect classical and solid work of image retrieval in academia and industry.

• Classical Local Feature
• Deep Learning Feature (Global Feature)
• Deep Learning Feature (Local Feature)
• Cross Model Retrieval (Cross Model Retrieval)
• ANN search
• CBIR rank
• CBIR in Industry
• CBIR Competition and Challenge
• CBIR for Duplicate(copy) detection
• Feature Fusion
• Instance Matching
• Semantic Matching
• Image Identification
• Tutorials
• Demo and Demo Online
• Datasets
• Useful Package

Classical Local Feature

• Object retrieval with large vocabularies and fast spatial matching
• Visual Categorization with Bags of Keypoints
• ORB: an efficient alternative to SIFT or SURF
• Object Recognition from Local Scale-Invariant Features
• Total Recall: Automatic Query Expansion with a Generative Feature Model for Object Retrieval
• Three things everyone should know to improve object retrieval
• On-the-fly learning for visual search of large-scale image and video datasets
• All about VLAD
• Aggregating localdescriptors into a compact image representatio
• More About VLAD: A Leap from Euclidean to Riemannian Manifolds
• Hamming embedding and weak geometric consistency for large scale image search
• Revisiting the VLAD image representation, project
• Improving the Fisher Kernel for Large-Scale Image Classification
• Image Classification with the Fisher Vector: Theory and Practice
• Democratic Diffusion Aggregation for ImageRetrieval
• A Vote-and-Verify Strategy for Fast Spatial Verification in Image Retrieval
• Triangulation embedding and democratic aggregation for image search
• Efficient Large-scale Image Search With a Vocabulary Tree, code

Deep Learning Feature (Global Feature)

• SOLAR: Second-Order Loss and Attention for Image Retrieval, arxiv 2020.
• A Benchmark on Tricks for Large-scale Image Retrieval，arxiv 2020。
• Learning with Average Precision: Training Image Retrieval with a Listwise Loss, ICCV 2019。
• MultiGrain: a unified image embedding for classes and instances, arxiv 2019.
• Deep Image Retrieval:Learning Global Representations for Image search
• End-to-end Learning of Deep Visual Representations for Image retrieval, DIR更详细的论文说明
• What Is the Best Practice for CNNs Applied to Visual Instance Retrieval?, 关于layer选取的问题
• Bags of Local Convolutional Features for Scalable Instance Search
• Faster R-CNN Features for Instance Search
• Cross-dimensional Weighting for Aggregated Deep Convolutional Features, project
• Class-Weighted Convolutional Features for Image Retrieval
• Multi-Scale Orderless Pooling of Deep Convolutional Activation Features, VLAD coding
• Aggregating Deep Convolutional Features for Image Retrieval, 论文笔记, 基于深度学习的视觉实例搜索研究进展.
• Particular object retrieval with integral max-pooling of CNN activations, project
• Particular object retrieval using CNN
• Learning to Match Aerial Images with Deep Attentive Architectures.
• Siamese Network of Deep Fisher-Vector Descriptors for Image Retrieval
• Combining Fisher Vector and Convolutional Neural Networks for Image Retrieval, fv和cnn特征融合提升
• Selective Deep Convolutional Features for Image Retrieval
• Class-Weighted Convolutional Features for Image Retrieval
• Towards Good Practices for Image Retrieval Based on CNN Features
• Fine-tuning CNN Image Retrieval with No Human Annotation
• An accurate retrieval through R-MAC+ descriptors for landmark recognition
• Regional Attention Based Deep Feature for Image Retrieval, code, BMVC 2018.
• Detect-to-Retrieve: Efficient Regional Aggregation for Image Search, arxiv.
• Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking, project, CVPR 2018.
• Guided Similarity Separation for Image Retrieval, NeurIPS 2019.

Deep Learning Feature (Local Feature)

• UR2KiD: Unifying Retrieval, Keypoint Detection, and Keypoint Description without Local Correspondence Supervision, arxiv.
• Visualizing Deep Similarity Networks.
• Combination of Multiple Global Descriptors for Image Retrieval.
• Beyond Cartesian Representations for Local Descriptors, code, ICCV 2019.
• R2D2: Reliable and Repeatable Detector and Descriptor, R2D2, NeurIPS 2019.
• SOSNet: Second Order Similarity Regularization for Local Descriptor Learning, CVPR 2019.
• Local Features and Visual Words Emerge in Activations, CVPR 2019.
• Explicit Spatial Encoding for Deep Local Descriptors, CVPR 2019.
• Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters, ICCV 2019.
• Learning Discriminative Affine Regions via Discriminability, affnet
• A Large Dataset for Improving Patch Matching, PS-Dataset
• Working hard to know your neighbor's margins: Local descriptor learning loss, hardnet
• MatchNet: Unifying Feature and Metric Learning for Patch-Based Matching, matchnet
• LF-Net: Learning Local Features from Images, NeurIPS 2018.
• Local Descriptors Optimized for Average Precision, CVPR 2018
• SuperPoint: Self-Supervised Interest Point Detection and Description, Magic Leap
• GeoDesc: Learning Local Descriptors by Integrating Geometry Constraints, code, ECCV 2018.
• Learning local feature descriptors with triplets and shallow convolutional neural networks, BMVC 2016.

Cross Model Retrieval

• Composing Text and Image for Image Retrieval - An Empirical Odyssey

ANN search

• RobustiQ A Robust ANN Search Method for Billion-scale Similarity Search on GPUs, ICMR 2019.
• Zoom: Multi-View Vector Search for Optimizing Accuracy, Latency and Memory
• Vector and Line Quantization for Billion-scale Similarity Search on GPUs
• GGNN: Graph-based GPU Nearest Neighbor Search, arxiv 2019.
• Learning to Route in Similarity Graphs, ICML 2019.
• Practical and Optimal LSH for Angular Distance
• pq-fast-scan
• faiss. A library for efficient similarity search and clustering of dense vectors.
• Polysemous codes
• Optimized Product Quantization
• lopq. Training of Locally Optimized Product Quantization (LOPQ) models for approximate nearest neighbor search of high dimensional data in Python and Spark.
• nns_benchmark. Benchmark of Nearest Neighbor Search on High Dimensional Data.
• Optimized Product Quantization
• Falconn. FAst Lookups of Cosine and Other Nearest Neighbors.
• Annoy. Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk
• NMSLIB. Non-Metric Space Library (NMSLIB): A similarity search library and a toolkit for evaluation of k-NN methods for generic non-metric spaces.
• Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs, graph-based method.
• Fast Approximate Nearest Neighbor Search With Navigating Spreading-out Graphs, code
• Efficient Nearest Neighbors Search for Large-Scale Landmark Recognition
• NV-tree: A Scalable Disk-Based High-Dimensional Index
• Dynamicity and Durability in Scalable Visual Instance Search
• Revisiting the Inverted Indices for Billion-Scale Approximate Nearest Neighbors，code
• Link and code: Fast indexing with graphs and compact regression codes
• A Survey of Product Quantization，对于矢量量化方法一篇比较完整的调研，值得一读
• GeoDesc: Learning Local Descriptors by Integrating Geometry Constraints，学习局部特征的descriptor，匹配能力较强
• Learning a Complete Image Indexing Pipeline, CVPR 2018
• spreading vectors for similarity search, ICLR 2019.
• SPTAG: A library for fast approximate nearest neighbor search. Microsoft.

CBIR rank

• Fast Spectral Ranking for Similarity Search, code, CVPR 2018

CBIR in Industry

• Videntifier is a visual search engine based on a patented large-scale local feature database, demo, based on SIFT feature and NV-tree.
• Web-Scale Responsive Visual Search at Bing
• Visual Search at Alibaba
• Visual Search at Pinterest
• Visual Discovery at Pinterest
• Learning a Unified Embedding for Visual Search at Pinterest, KDD 2019.
• Visual Search at ebay
• Deep Learning based Large Scale Visual Recommendation and Search for E-Commerce, project
• 微信「扫一扫识物」 的背后技术揭秘
• 揭秘微信「扫一扫」识物为什么这么快？

CBIR Competition and Challenge

• Google Landmark Retrieval Challenge, 2018
• Alibaba Large-scale Image Search Challenge, 2015
• Pkbigdata image retrieval, 2015
• Large-scale Landmark Retrieval/Recognition under a Noisy and Diverse Dataset, Landmark2019-1st-and-3rd-Place-Solution.

CBIR for Duplicate(copy) detection

• A Robust and Fast Video Copy Detection System Using Content-Based Fingerprinting

Feature Fusion

• Feature fusion using Canonical Correlation Analysis

Instance Matching

• Graph-Cut RANSAC, code
• Image Matching Benchmark
• GMS: Grid-based Motion Statistics for Fast, Ultra-robust Feature Correspondence
• A Vote-and-Verify Strategy for Fast Spatial Verification in Image Retrieval
• CODE: Coherence Based Decision Boundaries for Feature Correspondence
• Robust feature matching in 2.3µs
• PopSift is an implementation of the SIFT algorithm in CUDA
• openMVG robust_estimation
• Neural-Guided RANSAC: Learning Where to Sample Model Hypotheses.
• Homography from two orientation- and scale-covariant features, code.

Semantic Matching

• End-to-end weakly-supervised semantic alignment

Image Identification

• Image Identification Using SIFT Algorithm: Performance Analysis against Different Image Deformations

Tutorials

• How to Apply Distance Metric Learning to Street-to-Shop Problem
• Recent Image Search Techniques
• Compact Features for Visual Search
• multimedia-indexing. A framework for large-scale feature extraction, indexing and retrieval.
• Image Similarity using Deep Ranking, code.
• Triplet Loss and Online Triplet Mining in TensorFlow
• tf_retrieval_baseline.

Slide

• VRG Prague in “Large-Scale Landmark Recognition Challenge”, ranked 3rd in the Google Landmark Recognition Challenge.

Demo and Demo Online

• Visual Image Retrieval and Localization, SIFT feature encoded by BOW.
• VGG Image Search Engine, SIFT feature encoded by BOW.
• SoTu, A flask-based cbir system.
• yisou, A flask-based painting cbir system, the search algorithm is designed by Yong Yuan.

Datasets

• DeepFashion2 Dataset, DeepFashion2 is a comprehensive fashion dataset.

Useful Package

• VLFeat
• Yael