Sharing Matters for Generalization in Deep Metric Learning.

Learning the similarity between images constitutes the foundation for numerous vision tasks. The common paradigm is discriminative metric learning, which seeks an embedding that separates different training classes. However, the main challenge is to learn a metric that not only generalizes from training to novel, but related, test samples. It should also transfer to different object classes. So what complementary information is missed by the discriminative paradigm? Besides finding characteristics that separate between classes, we also need them to likely occur in novel categories, which is indicated if they are shared across training classes. This work investigates how to learn such characteristics without the need for extra annotations or training data. By formulating our approach as a novel triplet sampling strategy, it can be easily applied on top of recent ranking loss frameworks. Experiments show that, independent of the underlying network architecture and the specific ranking loss, our approach significantly improves performance in deep metric learning, leading to new the state-of-the-art results on various standard benchmark datasets.

VidTr: Video Transformer Without Convolutions.

We introduce Video Transformer (VidTr) with separable-attention for video classification. Comparing with commonly used 3D networks, VidTr is able to aggregate spatio-temporal information via stacked attentions and provide better performance with higher efficiency. We first introduce the vanilla video transformer and show that transformer module is able to perform spatio-temporal modeling from raw pixels, but with heavy memory usage. We then present VidTr which reduces the memory cost by 3.3× while keeping the same performance. To further optimize the model, we propose the standard deviation based topK pooling for attention (pooltopK_std), which reduces the computation by dropping non-informative features along temporal dimension. VidTr achieves state-of-the-art performance on five commonly used datasets with lower computational requirement, showing both the efficiency and effectiveness of our design. Finally, error analysis and visualization show that VidTr is especially good at predicting actions that require long-term temporal reasoning.

Early reduced behavioral activity induced by large strokes affects the efficiency of enriched environment in rats.

The majority of stroke patients develop post-stroke fatigue, a symptom which impairs motivation and diminishes the success of rehabilitative interventions. We show that large cortical strokes acutely reduce activity levels in rats for 1-2 weeks as a physiological response paralleled by signs of systemic inflammation. Rats were exposed early (1-2 weeks) or late (3-4 weeks after stroke) to an individually monitored enriched environment to stimulate self-controlled high-intensity sensorimotor training. A group of animals received Anti-Nogo antibodies for the first two weeks after stroke, a neuronal growth promoting immunotherapy already in clinical trials. Early exposure to the enriched environment resulted in poor outcome: Training intensity was correlated to enhanced systemic inflammation and functional impairment. In contrast, animals starting intense sensorimotor training two weeks after stroke preceded by the immunotherapy revealed better recovery with functional outcome positively correlated to the training intensity and the extent of re-innervation of the stroke denervated cervical hemi-cord. Our results suggest stroke-induced fatigue as a biological purposeful reaction of the organism during neuronal remodeling, enabling new circuit formation which will then be stabilized or pruned in the subsequent rehabilitative training phase. However, intense training too early may lead to wrong connections and is thus less effective.