ESPNetv2: A Light-Weight, Power Efficient, and General Purpose Convolutional Neural Network

Abstract

We introduce a light-weight, power efficient, and general purpose convolutional neural network, ESPNetv2, for modeling visual and sequential data. Our network uses group point-wise and depth-wise dilated separable convolutions to learn representations from a large effective receptive field with fewer FLOPs and parameters. The performance of our network is evaluated on four different tasks:(1) object classification, (2) semantic segmentation, (3) object detection, and (4) language modeling. Experiments on these tasks, including image classification on the Ima-geNet and language modeling on the PenTree bank dataset, demonstrate the superior performance of our method over the state-of-the-art methods. Our network outperforms ES-PNet by 4-5% and has 2 − 4× fewer FLOPs on the PASCAL VOC and the Cityscapes dataset. Compared to YOLOv2 on the MS-COCO object detection, ESPNetv2 delivers 4.4% higher accuracy with 6× fewer FLOPs. Our experiments show that ESPNetv2 is much more power efficient than existing state-of-the-art efficient methods including ShuffleNets and MobileNets. Our code is open-source and available at https://github.com/sacmehta/ ESPNetv2.

FDDWNet: A Lightweight Convolutional Neural Network for Real-Time Semantic Segmentation

“…Standard [21] n 2 cĉ n × n Group [15] n 2 cĉ/g n × n 1D-factorized [13] 2ncĉ n × n DW [11,12] n 2 c + cĉ n × n DDW [14] n 2 c + cĉ n r × n r our FDDWC 2nc + cĉ n r × n r…”

“…Towards this end, this section introduces EERM, the core unit of FDDWNet, to approach the representational power of larger and denser layers, but at a considerably lower computational budgets. EERM unit, which leverages the residual connections and FDDWC, combines the strength of 1D-factorized convolution [13] and dilated depth-wise separable convolution [14]. More specifically, as shown in Fig.…”

“…ERFNet [13] decomposes a 2D convolution (e.g., 3 × 3) into two 1D-factorized convolution (e.g., 3 × 1 and 1 × 3). An alternative efficient approach to lighten CNNs depends on group convolution [14,15], where input channels and filter kernels are accordingly factored into a set of groups and each group is convolved independently. In spite of achieving impressive results, these lightweight networks prefer to adopt shallow network architectures to reduce model complexity, which may weaken the representation ability of visual data, leading to the degradation of performance.…”

“…In this paper, we introduce a novel lightweight network, called FDDWNet, for the task of real-time semantic segmentation. Different from previous methods [11,14,16,17,18], our FDDWNet makes an effort to design more deeper network architecture to develop the ability of feature representation, while maintains very fewer model parameters to accelerate inference speed. In addition, FDDWNet has multiple branches of skipped connections from intermediate convolution layers, in spite of adding a bit of computational burden, but helping to gather more context.…”

Depthwise-STFT Based Separable Convolutional Neural Networks

“…Recently, there has been a growing interest into developing space-time efficient neural networks for real time and resource restricted applications [10,6,7,9,8,11,12]. Depthwise Separable Convolutions.…”