Leveraging category-level labels for instance-level image retrieval

Gordoa, Albert; Rodríguez-Serrano, José A.; Perronnin, Florent; Valveny, Ernest

doi:10.1109/cvpr.2012.6248035

Cited by 61 publications

(51 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using classifiers to learn a metric for image retrieval was recently considered also in [16]. They found the Joint Subspace and Classifier Learning (JSCL) method to be most effective.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…The performance is measured by 4×recall@4 averaged over all queries, hence the maximal score is 4. For both datasets we extract the 4K image features also used in our earlier experiments, which are the same ones as those used in [16]. To compute the Table 7: Instance level image retrieval accuracy on the Holidays dataset, and the UKB dataset.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…For this experiment we use the same public benchmarks as in [16]. First, the IN-RIA Holidays data set introduced by [20] consists of 1,491 images of 500 scenes and objects.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…the dot-product on 2 -normalized vectors. In analogy to [16] we use NCMML to train a metric from the ILSVRC'10 data set, and do the early stopping based on retrieval performance. To avoid tuning on the test data the cross-validation is performed on the other dataset, i.e.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…To compute the Table 7: Instance level image retrieval accuracy on the Holidays dataset, and the UKB dataset. NCMML is compared to a PCA baseline and the JSCL method [16].…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

See 4 more Smart Citations

Large Scale Metric Learning for Distance-Based Image Classification on Open Ended Data Sets

Mensink¹,

Verbeek²,

Perronnin³

2013

Advanced Topics in Computer Vision

View full text Add to dashboard Cite

Many real-life large-scale datasets are open-ended and dynamic: new images are continuously added to existing classes, new classes appear over time, and the semantics of existing classes might evolve too. Therefore, we study large-scale image classification methods that can incorporate new classes and training images continuously over time at negligible cost. To this end we consider two distancebased classifiers, the k-nearest neighbor (k-NN) and nearest class mean (NCM) classifiers. Since the performance of distance-based classifiers heavily depends on the used distance function, we cast the problem into one of learning a low-rank metric, which is shared across all classes. For the NCM classifier we introduce a new metric learning approach, and we also introduce an extension to allow for richer class representations.Experiments on the ImageNet 2010 challenge dataset, which contains over one million training images of thousand classes, show that, surprisingly, the NCM classifier compares favorably to the more flexible k-NN classifier. Moreover, the NCM performance is comparable to that of linear SVMs which obtain current state-of-theart performance. Experimentally we study the generalization performance to classes that were not used to learn the metrics. Using a metric learned on 1,000 classes, we show results for the ImageNet-10K dataset which contains 10,000 classes, and obtain performance that is competitive with the current state-of-the-art, while being orders of magnitude faster.

show abstract

“…Using classifiers to learn a metric for image retrieval was recently considered also in [16]. They found the Joint Subspace and Classifier Learning (JSCL) method to be most effective.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…For this experiment we use the same public benchmarks as in [16]. First, the IN-RIA Holidays data set introduced by [20] consists of 1,491 images of 500 scenes and objects.…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

Section: Instance Level Image Retrievalmentioning

confidence: 99%

“…To compute the Table 7: Instance level image retrieval accuracy on the Holidays dataset, and the UKB dataset. NCMML is compared to a PCA baseline and the JSCL method [16].…”

Section: Instance Level Image Retrievalmentioning

confidence: 99%

See 3 more Smart Citations