Fine-grained visual categorization (FGVC) is a challenging task because there are many hard examples existing between fine-grained classes which differ subtly in particular local regions. To address this issue, many methods have recourse to high-resolution source images and others adopt effective regularization like "mixup" or "between class learning." Despite their promising achievements, mixup tends to cause the manifold intrusion problem which would result in under-fitting and degradation of the model performance and high-resolution input inevitably leads to high computational costs. In view of this, we present a multiresolution discriminative mixup network (MRDMN). Different from standard mixup, the proposed discriminative mixup strategy mixes discriminative regions linearly instead of entire images to avoid manifold intrusion, which makes it learn the local detail features more effectively and contributes to more precise categorization. Furthermore, an innovative resolution-based distillation strategy is designed to transfer the multiresolution detail feature representations to a low-resolution network, which speeds up the testing and boosts the categorization accuracy simultaneously. Extensive experiments demonstrate that our proposed MRDMN remarkably outperforms most competitive approaches with less computation time on the CUB-200-2011, Stanford-Cars, Stanford-Dogs, Food-101, and iNaturalist 2017 datasets. The codes are in https://github.com/aztc/MRDMN.

Orthogonality regularization has proven effective in improving the precision, convergence speed and the training stability of CNNs. Here, we propose a novel Orthogonal Dictionary Convolution Strategy (ODCS) on CNNs to improve orthogonality effect by optimizing the network architecture and changing the regularized object. Specifically, we remove the nonlinear layer in typical convolution block “Conv(BN) + Nonlinear + Pointwise Conv(BN)”, and only impose orthogonal regularization on the front Conv. The structure, “Conv(BN) + Pointwise Conv(BN)”, is then equivalent to a pair of dictionary and encoding, defined in sparse dictionary learning. Thanks to the exact and efficient representation of signal with dictionaries in low-dimensional projections, our strategy could reduce the superfluous information in dictionary Conv kernels. Meanwhile, the proposed strategy relieves the too strict orthogonality regularization in training, which makes hyper-parameters tuning of model to be more flexible. In addition, our ODCS can modify the state-of-the-art models easily without any extra consumption in inference phase. We evaluate it on a variety of CNNs in small-scale (CIFAR), large-scale (ImageNet) and fine-grained (CUB-200-2011) image classification tasks, respectively. The experimental results show that our method achieve a stable and superior improvement.

There are many AI applications in high-income countries because their implementation depends on expensive GPU cards (~2000$) and reliable power supply (~200W). To deploy AI in resource-poor settings on cheaper (~20$) and low-power devices (