Towards Backward-Compatible Representation Learning
CVPR, pp. 6367-6376, 2020.
The first is the accuracy gap of the new models trained with backward-compatible training relative to the new model oblivious of previous constraints
We propose a way to learn visual features that are compatible with previously computed ones even when they have different dimensions and are learned via different neural network architectures and loss functions. Compatible means that, if such features are used to compare images, then "new" features can be compared directly to "old" feat...More
PPT (Upload PPT)