Statistically Motivated Second Order Pooling
This addresses the deployment challenges of second-order networks for researchers and practitioners in computer vision, offering a more efficient solution.
The paper tackles the memory-intensive nature of second-order pooling in deep learning for visual recognition by introducing a parametric compression strategy that produces more compact representations than existing techniques, outperforming both compressed and uncompressed second-order models on benchmark datasets.
Second-order pooling, a.k.a.~bilinear pooling, has proven effective for deep learning based visual recognition. However, the resulting second-order networks yield a final representation that is orders of magnitude larger than that of standard, first-order ones, making them memory-intensive and cumbersome to deploy. Here, we introduce a general, parametric compression strategy that can produce more compact representations than existing compression techniques, yet outperform both compressed and uncompressed second-order models. Our approach is motivated by a statistical analysis of the network's activations, relying on operations that lead to a Gaussian-distributed final representation, as inherently used by first-order deep networks. As evidenced by our experiments, this lets us outperform the state-of-the-art first-order and second-order models on several benchmark recognition datasets.