Toward Generalized Visualization System for Various CNN Models: Module and Layer View

Yunsu Kim, Junha Chun, Sungwook Son, Hyunwoo Park

Abstract: We present CNNExplorer, a generalized visualization system designed to assist beginners in understanding Convolutional Neural Networks (CNNs). CNNExplorer provides an intuitive interface and detailed visualizations at both the module and layer levels, enabling users to explore and understand the complex structures and operations of CNN models without requiring extensive coding knowledge. The module view abstracts the complexity by grouping layers into modules, while the layer view shows detailed transformations within each module. Our system supports a wide range of CNN models, including those available on Hugging Face, allowing users to visualize and analyze a variety of pre-trained models. To demonstrate the potential effectiveness of CNNExplorer in practical applications, we proposed specific application scenarios. Furthermore, we conducted a user study with participants and found the system is effective in enhancing their understanding of deep learning models and appreciated the interactive features. In future work, we will focus on supporting the latest vision models and expanding visualization capabilities to other deep learning architectures.

External IDs:dblp:journals/access/KimCSP24