MixBCT: Towards Self-Adapting Backward-Compatible Training
This work addresses a practical issue for retrieval system developers by enabling efficient model updates without database re-indexing, though it appears incremental as it builds on prior prototype and instance-based methods.
The paper tackles the problem of backward-compatible training in retrieval systems, where deploying new models without updating old databases is costly, and proposes MixBCT, a method that adaptively adjusts constraints based on old feature distributions, achieving superior performance on MS1Mv3 and IJB-C face recognition datasets.
Backward-compatible training circumvents the need for expensive updates to the old gallery database when deploying an advanced new model in the retrieval system. Previous methods achieved backward compatibility by aligning prototypes of the new model with the old one, yet they often overlooked the distribution of old features, limiting their effectiveness when the low quality of the old model results in a weakly feature discriminability. Instance-based methods like L2 regression take into account the distribution of old features but impose strong constraints on the performance of the new model itself. In this paper, we propose MixBCT, a simple yet highly effective backward-compatible training method that serves as a unified framework for old models of varying qualities. We construct a single loss function applied to mixed old and new features to facilitate backward-compatible training, which adaptively adjusts the constraint domain for new features based on the distribution of old features. We conducted extensive experiments on the large-scale face recognition datasets MS1Mv3 and IJB-C to verify the effectiveness of our method. The experimental results clearly demonstrate its superiority over previous methods. Code is available at https://github.com/yuleung/MixBCT .