Deep neural networks have been utilized in an increasing number of computer vision tasks, demonstrating superior performance. Much research has been focused on making deep networks more suitable for efficient hardware implementation, for low-power and low-latency real-time applications. In [1], Isikdogan et al. introduced a deep neural network design that provides an effective trade-off between flexibility and hardware efficiency. The proposed solution consists of fixed-topology hardware blocks, with partially frozen/partially trainable weights, that can be configured into a full network. Initial results in a few computer vision tasks were presented in [1]. In this paper, we further evaluate this network design by applying it to several additional computer vision use cases and comparing it to other hardware-friendly networks. The experimental results presented here show that the proposed semi-fixed semi-frozen design achieves competitive performanc on a variety of benchmarks, while maintaining very high hardware efficiency.
We present a practical 3D-assited face alignment framework based on cascaded regression in this paper. The 3D information embedded in 2D face image is utilized to calculate two novel components to improve the performance of 2D methods in unconstrained face alignment. The two novel components for 2D image features are the projected local patch and the visibility of each landmark. First, we propose to extract the landmark related features in the projected local patches on 2D image from the corresponding 3D face model. Local patches of a fixed landmark in 3D face models for different 2D images cover the same region of face anatomically. The extracted features are more accurate for further locations regression of landmarks. Second, we propose to estimate the visibilities of 2D landmarks based on 3D face model, which are proven to be vital to address large pose face alignment problem. In this paper, we adopt Local Binary Features (LBF) to extract landmark related features in the proposed framework, and name the new method as 3D-Assisted LBF (3DALBF). An extensive evaluation on two face databases shows that 3DALBF can achieve better alignment results than the original 2D method and maintain the speed advantage of 2D method over 3D method.