Unsupervised Data Uncertainty Learning in Visual Retrieval Systems
This work addresses the problem of noisy data and interpretability in visual retrieval systems for researchers and practitioners, though it appears incremental as it builds on existing triplet loss methods.
The paper tackles the problem of estimating data uncertainty in visual retrieval systems by introducing an unsupervised formulation that extends triplet loss to model heteroscedastic uncertainty for each input, improving performance and interpretability. Results show utility in image and video retrieval applications, with evaluations on Clothing1M and Honda Driving datasets demonstrating efficiency in modeling local noise and identifying confusing scenarios.
We introduce an unsupervised formulation to estimate heteroscedastic uncertainty in retrieval systems. We propose an extension to triplet loss that models data uncertainty for each input. Besides improving performance, our formulation models local noise in the embedding space. It quantifies input uncertainty and thus enhances interpretability of the system. This helps identify noisy observations in query and search databases. Evaluation on both image and video retrieval applications highlight the utility of our approach. We highlight our efficiency in modeling local noise using two real-world datasets: Clothing1M and Honda Driving datasets. Qualitative results illustrate our ability in identifying confusing scenarios in various domains. Uncertainty learning also enables data cleaning by detecting noisy training labels.