Fine-tuning and visualization of convolutional neural networks
Xiangnan Yin, Weihai Chen, Xingming Wu, Haosong Yue · 2017
Image classification is a widely discussed topic in the field of computer vision. In recent years, with the application of Convolutional Neural Networks (CNNs), the state-of-the-art in this area has progressed rapidly. To yield a well performed CNN, the advanced GPU and large amount of training data are employed, thus training an entire CNN from scratch is difficult. In practice, fine-tuning a pre-trained CNN is a simple yet effective method to solve a target task. In this paper, we address on the issue of visualizing a fine-tuned CNN, comparing with a small CNN trained from scratch on the same task, to explain how fine-tuning achieve such good performance.