TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/Recall@k Surrogate Loss with Large Batches and Similarity ...

Recall@k Surrogate Loss with Large Batches and Similarity Mixup

Yash Patel, Giorgos Tolias, Jiri Matas

2021-08-25CVPR 2022 1Vehicle Re-IdentificationMetric LearningRetrievalImage Retrieval
PaperPDFCode(official)Code(official)

Abstract

This work focuses on learning deep visual representation models for retrieval by exploring the interplay between a new loss function, the batch size, and a new regularization approach. Direct optimization, by gradient descent, of an evaluation metric, is not possible when it is non-differentiable, which is the case for recall in retrieval. A differentiable surrogate loss for the recall is proposed in this work. Using an implementation that sidesteps the hardware constraints of the GPU memory, the method trains with a very large batch size, which is essential for metrics computed on the entire retrieval database. It is assisted by an efficient mixup regularization approach that operates on pairwise scalar similarities and virtually increases the batch size further. The suggested method achieves state-of-the-art performance in several image retrieval benchmarks when used for deep metric learning. For instance-level recognition, the method outperforms similar approaches that train using an approximation of average precision.

Results

TaskDatasetMetricValueModel
Image RetrievaliNaturalistR@183Recall@k Surrogate loss (ViT-B/16)
Image RetrievaliNaturalistR@1695.9Recall@k Surrogate loss (ViT-B/16)
Image RetrievaliNaturalistR@3297.2Recall@k Surrogate loss (ViT-B/16)
Image RetrievaliNaturalistR@592.1Recall@k Surrogate loss (ViT-B/16)
Image RetrievaliNaturalistR@171.8Recall@k Surrogate loss (ResNet-50)
Image RetrievaliNaturalistR@1691.9Recall@k Surrogate loss (ResNet-50)
Image RetrievaliNaturalistR@3294.3Recall@k Surrogate loss (ResNet-50)
Image RetrievaliNaturalistR@584.7Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID LargeRank-194.7Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID LargeRank-597.1Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID LargeRank-193.8Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID LargeRank-596.6Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID MediumRank-195.2Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID MediumRank-597.2Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID MediumRank-194.6Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID MediumRank-596.9Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID SmallRank-196.2Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID SmallRank-598Recall@k Surrogate loss (ViT-B/16)
Intelligent SurveillanceVehicleID SmallRank-195.7Recall@k Surrogate loss (ResNet-50)
Intelligent SurveillanceVehicleID SmallRank-597.9Recall@k Surrogate loss (ResNet-50)
Metric LearningCARS196R@189.5Recall@k Surrogate loss (ViT-B/16)
Metric LearningCARS196R@188.3Recall@k Surrogate loss (ResNet-50)
Metric LearningStanford Online ProductsR@188Recall@k Surrogate Loss (ViT-B/16)
Metric LearningStanford Online ProductsR@185.1Recall@k Surrogate Loss (ViT-B/32)
Metric LearningStanford Online ProductsR@182.7Recall@k Surrogate Loss (ResNet-50)
Vehicle Re-IdentificationVehicleID LargeRank-194.7Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID LargeRank-597.1Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID LargeRank-193.8Recall@k Surrogate loss (ResNet-50)
Vehicle Re-IdentificationVehicleID LargeRank-596.6Recall@k Surrogate loss (ResNet-50)
Vehicle Re-IdentificationVehicleID MediumRank-195.2Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID MediumRank-597.2Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID MediumRank-194.6Recall@k Surrogate loss (ResNet-50)
Vehicle Re-IdentificationVehicleID MediumRank-596.9Recall@k Surrogate loss (ResNet-50)
Vehicle Re-IdentificationVehicleID SmallRank-196.2Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID SmallRank-598Recall@k Surrogate loss (ViT-B/16)
Vehicle Re-IdentificationVehicleID SmallRank-195.7Recall@k Surrogate loss (ResNet-50)
Vehicle Re-IdentificationVehicleID SmallRank-597.9Recall@k Surrogate loss (ResNet-50)

Related Papers

Unsupervised Ground Metric Learning2025-07-17From Roots to Rewards: Dynamic Tree Reasoning with RL2025-07-17HapticCap: A Multimodal Dataset and Task for Understanding User Experience of Vibration Haptic Signals2025-07-17A Survey of Context Engineering for Large Language Models2025-07-17MCoT-RE: Multi-Faceted Chain-of-Thought and Re-Ranking for Training-Free Zero-Shot Composed Image Retrieval2025-07-17FAR-Net: Multi-Stage Fusion Network with Enhanced Semantic Alignment and Adaptive Reconciliation for Composed Image Retrieval2025-07-17Are encoders able to learn landmarkers for warm-starting of Hyperparameter Optimization?2025-07-16Developing Visual Augmented Q&A System using Scalable Vision Embedding Retrieval & Late Interaction Re-ranker2025-07-16