Efficient Yet Deep Convolutional Neural Networks for Semantic Segmentation

TL;DR

This paper introduces two efficient deep convolutional neural network architectures for semantic segmentation that require fewer parameters, achieve higher accuracy, and are faster and less memory-intensive than existing models.

Contribution

The authors propose novel segmentation architectures based on transfer learning, dilated convolutions, and skip connections, reducing parameter count and improving performance over similar models.

Findings

Models use one-third the parameters of comparable architectures.

Achieve better accuracy on Pascal VOC2012, Pascal-Context, and NYUDv2 datasets.

Faster inference time and lower memory consumption on NVIDIA Pascal GPUs.

Abstract

Semantic Segmentation using deep convolutional neural network pose more complex challenge for any GPU intensive task. As it has to compute million of parameters, it results to huge memory consumption. Moreover, extracting finer features and conducting supervised training tends to increase the complexity. With the introduction of Fully Convolutional Neural Network, which uses finer strides and utilizes deconvolutional layers for upsampling, it has been a go to for any image segmentation task. In this paper, we propose two segmentation architecture which not only needs one-third the parameters to compute but also gives better accuracy than the similar architectures. The model weights were transferred from the popular neural net like VGG19 and VGG16 which were trained on Imagenet classification data-set. Then we transform all the fully connected layers to convolutional layers and use dilated convolution for decreasing the parameters. Lastly, we add finer strides and attach four skip architectures which are element-wise summed with the deconvolutional layers in steps. We train and test on different sparse and fine data-sets like Pascal VOC2012, Pascal-Context and NYUDv2 and show how better our model performs in this tasks. On the other hand our model has a faster inference time and consumes less memory for training and testing on NVIDIA Pascal GPUs, making it more efficient and less memory consuming architecture for pixel-wise segmentation.