Caffe2Unity: Immersive Visualization and Interpretation of Deep Neural Networks

Abstract

Deep neural networks (DNNs) dominate many tasks in the computer vision domain, but it is still difficult to understand and interpret the information contained within these networks. To gain better insight into how a network learns and operates, there is a strong need to visualize these complex structures, and this remains an important research direction. In this paper, we address the problem of how the interactive display of DNNs in a virtual reality (VR) setup can be used for general understanding and architectural assessment. We compiled a static library as a plugin for the Caffe framework in the Unity gaming engine. We used routines from this plugin to create and visualize a VR-based AlexNet architecture for an image classification task. Our layered interactive model allows the user to freely navigate back and forth within the network during visual exploration. To make the DNN model even more accessible, the user can select certain connections to understand the activity flow at a particular neuron. Our VR setup also allows users to hide the activation maps/filters or even interactively occlude certain features in an image in real-time. Furthermore, we added an interpretation module and reframed the Shapley values to give a deeper understanding of the different layers. Thus, this novel tool offers more direct access to network structures and results, and its immersive operation is especially instructive for both novices and experts in the field of DNNs.

Interpreting the decisions of CNNs via influence functions

“…To further evaluate and interpret the learned representations, we identified Shapley value-based (Lundberg and Lee, 2017) influential regions between different types of test inputs (Figure 10). Here, we did not calculate layer-wise Shapley values but only considered the test images to see which image regions were important for the network using our previous work (Aamir et al, 2022). The reason for this analysis is that we wanted to identify what the network looks at in making its decision.…”

Visualization of AI Systems in Virtual Reality: A Comprehensive Review

“…Unity, as a gaming engine, was a common choice among researchers. For instance, [21] used Unity to create a virtual reality environment for visualizing deep neural networks built using the Caffe framework.The research [22] and [23] both utilized Unity in their research, illustrating its dual functionality. They used Unity for creating virtual reality environments and developing deep convolutional neural network models.…”

A walk in the black-box: 3D visualization of large neural networks in virtual reality

“…Aamir et al [10] presented a novel approach to immersively visualize and interpret deep networks in VR, where the user can move freely inside an AlexNet [28]. The layers are represented as a sequence of 2D planes in 3D space showing the activations, which we adopted in our approach.…”

“…• Rotate the image within the range of [1,10] degrees with a probability of 30%. • Scale the image with a factor from range [1., 1.05] and a probability of 5%.…”

“…Displaying CNNs in 3D environments in virtual reality (VR) is a promising approach to make CNNs more humanunderstandable and accessible [6]. A handful of prototypes [7][8][9][10] show that VR can accelerate the general understanding of CNNs and offer an intuitive way to interact with these complex structures. The immersive experience provides new possibilities for CNN architecture assessment.…”