Large-to-small Image Resolution Asymmetry in Deep Metric Learning
This work addresses efficiency in image retrieval for applications like visual search, but it is incremental as it builds on prior asymmetry methods by focusing on resolution instead of architecture.
The paper tackles the performance/efficiency trade-off in deep metric learning by proposing an asymmetric setup where database images use high resolution for fine details and query images use low resolution for fast processing, achieving competitive results on standard benchmarks like CUB200, Cars196, and SOP.
Deep metric learning for vision is trained by optimizing a representation network to map (non-)matching image pairs to (non-)similar representations. During testing, which typically corresponds to image retrieval, both database and query examples are processed by the same network to obtain the representation used for similarity estimation and ranking. In this work, we explore an asymmetric setup by light-weight processing of the query at a small image resolution to enable fast representation extraction. The goal is to obtain a network for database examples that is trained to operate on large resolution images and benefits from fine-grained image details, and a second network for query examples that operates on small resolution images but preserves a representation space aligned with that of the database network. We achieve this with a distillation approach that transfers knowledge from a fixed teacher network to a student via a loss that operates per image and solely relies on coupled augmentations without the use of any labels. In contrast to prior work that explores such asymmetry from the point of view of different network architectures, this work uses the same architecture but modifies the image resolution. We conclude that resolution asymmetry is a better way to optimize the performance/efficiency trade-off than architecture asymmetry. Evaluation is performed on three standard deep metric learning benchmarks, namely CUB200, Cars196, and SOP. Code: https://github.com/pavelsuma/raml