We propose to use hyperbolic uncertainty to determine the algorithmic learning pace, under the assumption that less uncertain samples should be more strongly driving the training, with a larger weight and pace.
Ranked #57 on Skeleton Based Action Recognition on NTU RGB+D 120
By using uncertainty, HypAD may assess whether it is certain about the input signal but it fails to reconstruct it because this is anomalous; or whether the reconstruction error does not necessarily imply anomaly, as the model is uncertain, e. g. a complex but regular input signal.
Few-shot fine-grained classification and person search appear as distinct tasks and literature has treated them separately.
In experimental evaluation, the combination of CIR and a plain Siamese-net with triplet loss yields best few-shot learning performance on the challenging tieredImageNet.
Notably, our joint optimization maintains the detector performance, a typical multi-task challenge.
We employ this to supervise the detector of our person search model at various levels using a specialized detector.
We extend this with i. a query-guided Siamese squeeze-and-excitation network (QSSE-Net) that uses global context from both the query and gallery images, ii.