Search

Pseudo-supervised learning methods have been shown to be effective for weakly supervised object localization tasks. However, the effectiveness depends on the powerful regularization ability of deep neural networks. Based on the assumption that the localization network should have similar location predictions on different versions of the same image, we propose a two-stage approach to learn more consistent localization. In the first stage, we propose a mask-based pseudo label generator algorithm, and use the pseudo-supervised learning method to initialize an object localization network. In the second stage, we propose a simple and effective method for evaluating the confidence of pseudo-labels based on classification discrimination, and by learning consistency from high-quality pseudo-labels, we further refine the localization network to get better localization performance. Experimental results show that our proposed approach achieves excellent performance in three benchmark datasets including CUB-200-2011, ImageNet-1k and Tiny-ImageNet, which demonstrates its effectiveness.

Based on a subtle gesture video dataset,a deep network was proposed to improve the performance on such fine-grained actions in this paper.The architecture consisted of a smaller variation of convolutional 3-dimensional (C3D) network and a long short-term memory (LSTM) with a soft attention mechanism.The depth of C3D network and the weight penalty of the attention mechanism were optimized.Experimental results showed that the fine-grained action recognition network could effectively focus on the certain part with important information,and performed better on both average accuracy and detection accuracy for the dataset.