Visual Instance Retrieval with Deep Convolutional Networks

By Ali Sharif Razavian, Josephine Sullivan, Atsuto Maki, Stefan Carlsson. International Conference on Learning Representations, 2014

383 citations recorded by the source corpus.

This paper provides an extensive study on the availability of image representations based on convolutional networks (ConvNets) for the task of visual instance retrieval. Besides the choice of convolutional layers, we present an efficient pipeline exploiting multi-scale schemes to extract local features, in particular, by taking geometric invariance into explicit account, i.e. positions, scales and spatial consistency. In our experiments using five standard image retrieval datasets, we demonstrate that generic ConvNet image representations can outperform other state-of-the-art methods if they are extracted appropriately.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/323d6a6619b982f282acc5a27f142dc3cef7b2bd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Yi Yang
2. Luc Van Gool
3. Qi Tian
4. Dacheng Tao
5. Ming-Hsuan Yang
6. Liang Zheng
7. Xiaogang Wang
8. Trevor Darrell

Citations in the corpus

• Cross-Dimensional Weighting for Aggregated Deep Convolutional Features (2015, methodology): “[28, 29] proposed different pooling strategies for such features and demonstrated state-of-the-art performance in popular benchmarks for compact image representations, i.”
• Fine-Tuning CNN Image Retrieval with No Human Annotation (2017, methodology): “successfully perform image search [8], [9], [10], [11], [12].”
• An Evaluation Metric for Object Detection Algorithms in Autonomous Navigation Systems and its Application to a Real-Time Alerting System (2018, methodology): “The deep features of the image obtained from the conv5 1 [15] of the VGG net, called φR are compared with the subset of images of Virtual KITTI dataset [6] of the corresponding weather condition as predicted by the Random Forest.”
• Asymmetric metric learning for knowledge transfer (2020, background): “Convolutional networks quickly outperformed shallow representations, using different pooling mechanisms [56, 68] and fine-tuning on relevant datasets, initially with cross-entropy on noisy labels from the web [2] and then with contrastive [53] and triplet [16] loss on labels generated from the visua”
• Mining on Manifolds: Metric Learning Without Labels (2018, background): “Transfer learning from category-level classification is one case [43].”
• Efficient Indexing of Billion-Scale Datasets of Deep Descriptors (2016, background): “Several works [17, 7, 11, 18, 4] have shown that DNN-based descriptors significantly improve the state-of-the-art on common retrieval benchmarks.”
• Aggregating Local Deep Features for Image Retrieval (2015, methodology): “SPoC features provide considerable improvement over previous state-of-the-art for compact descriptors including deep descriptors in [2, 7, 22].”
• Exploiting local features from deep networks for image retrieval (2015, methodology): “It is surprising that the networks perform better with larger input images, although by default they should take images of 224 x 224 pixels that they are trained on as the input [26].”
• PatternNet: A Benchmark Dataset for Performance Evaluation of Remote Sensing Image Retrieval (2017, methodology): “Currently, CNNs have been widely used for various retrieval tasks ranging from computer vision to remote sensing [47,48,105–114].”
• Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models (2022, background): “Image retrieval works with image descriptors based on all types of neural networks [3, 53].”
• Explicit Shape Encoding for Real-Time Instance Segmentation (2019, background): “Recently, a number of works encode the shape implicitly [9, 29, 38], which is to project the shape content to a latent vector, typically through a black-box design such as deep CNN.”
• Web-Scale Image Clustering Revisited (2015, methodology): “Although several methods exist for aggregating local descriptors [17], we rather choose deep learned features, which are shown to be superior in low dimensions [5],[28].”
• Dynamic Match Kernel With Deep Convolutional Features for Image Retrieval (2018, neutral): “3† Razavian 2015 [39] N/A N/A N/A N/A 58.”
• SIFT Meets CNN: A Decade Survey of Instance Retrieval (2016, methodology): “Apart from the above-mentioned strategies, we notice that several works [7], [133], [152] extract several region descriptors per image to do a many-to-many matching, called “spatial search” [7].”
• Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking (2018, methodology): “These are fine-tuned ResNet with GeM pooling [37], pre-trained (on ImageNet) AlexNet with MAC pooling [38], and ASMK [46].”
• Polisis: Automated Analysis and Presentation of Privacy Policies Using Deep Learning (2018, result): “This approach is similar to what has been applied previously in image retrieval, where image representations learned from a large-scale image classification task were effective in visual search applications [50].”