Conventional model upgrades for visual search systems require offline refresh of gallery features by feeding gallery images into new models (dubbed as "backfill"), which is time-consuming and expensive, especially in large-scale applications. The task of backward-compatible representation learning is therefore introduced to support backfill-free model upgrades, where the new query features are interoperable with the old gallery features. Despite the success, previous works only investigated a close-set training scenario (i.e., the new training set shares the same classes as the old one), and are limited by more realistic and challenging open-set scenarios. To this end, we first introduce a new problem of universal backward-compatible representation learning, covering all possible data split in model upgrades. We further propose a simple yet effective method, dubbed as Universal Backward-Compatible Training (UniBCT) with a novel structural prototype refinement algorithm, to learn compatible representations in all kinds of model upgrading benchmarks in a unified manner. Comprehensive experiments on the large-scale face recognition datasets MS1Mv3 and IJB-C fully demonstrate the effectiveness of our method.
MixBCT: Towards Self-Adapting Backward-Compatible Training
Hunan University（湖南大学）；Intellifusion Inc.（云天励飞公司）；Peking University（北京大学）；Peng Cheng Laboratory（鹏城实验室）；The Hong Kong University of Science and Technology(Guangzhou)（香港科技大学（广州））
$BT^2$: Backward-compatible Training with Basis Transformation
University of California, Berkeley（加州大学伯克利分校）；Cornell University（康奈尔大学）；University of Maryland, College Park（马里兰大学帕克分校）；University of Hong Kong（香港大学）；MIT（麻省理工学院）；Meta AI
A Forward and Backward Compatible Framework for Few-shot Class-incremental Pill Recognition
National University of Defense Technology (NUDT)（国防科技大学）；College of Intelligence Science and Technology（智能科学与技术学院）；College of Electronic Science and Technology（电子科学与技术学院）
Improving Forward Compatibility in Class Incremental Learning by Increasing Representation Rank and Feature Richness
LINE Investment Technologies（LINE投资技术公司）；Seoul National University（首尔大学）；LG AI Research（LG人工智能研究院）；Research Institute for Convergence Science, Seoul National University（首尔大学融合科学研究院）
COCO等基准在目标检测中起着至关重要的作用。然而，现有的基准是不够的规模变化，其协议是不够的公平比较。在本文中，我们介绍了通用尺度目标检测基准（USB）。USB通过将COCO与最近提出的Waymo Open Dataset和Manga 109-s数据集相结合，在对象尺度和图像域上有所变化。为了实现公平比较和包容性研究，我们提出了培训和评估协议。他们有多个训练时期和评估图像分辨率的部门，如体育运动中的重量等级，以及跨训练协议的兼容性，如通用串行总线的向后兼容性。具体来说，我们要求参与者不仅报告更高协议（更长的培训）的结果，而且报告更低协议（更短的培训）的结果。使用提出的基准和协议，我们进行了广泛的实验，使用15种方法，发现现有的COCO偏置方法的弱点。该代码可在https://github.com/shinya7y/UniverseNet上获得。
