VisualBackProp: visualizing CNNs for autonomous driving

By Mariusz Bojarski, Anna Choromanska, Krzysztof Marcin Choromanski, Bernhard Firner, L. Jackel, Urs Muller, Karol Zieba. 2016

78 citations recorded by the source corpus.

This paper proposes a new method, that we call VisualBackProp, for visualizing which sets of pixels of the input image contribute most to the predictions made by the convolutional neural network (CNN). The method heavily hinges on exploring the intuition that the feature maps contain less and less irrelevant information to the prediction decision when moving deeper into the network. The technique we propose was developed as a debugging tool for CNN-based systems for steering self-driving cars and is therefore required to run in real-time, i.e. it was designed to require less computations than a forward propagation. This makes the presented visualization method a valuable debugging tool which can be easily used during both training and inference. We furthermore justify our approach with theoretical arguments and theoretically confirm that the proposed method identifies sets of input pixels, rather than individual pixels, that collaboratively contribute to the prediction. Our theoretical findings stand in agreement with the experimental results. The empirical evaluation shows the plausibility of the proposed approach on the road video data and reveals that it compares favorably to the LRP approach, i.e. it obtains similar visualization results and simultaneously achieves order of magnitude speed-ups.

Subfields: Automotive Engineering, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Automotive Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

A Survey of Methods for Explaining Black Box Models (2018, methodology): “Another approach f that can be used to solve both the outcome explanation problem and the model inspection in case of image classification with CNN is VisualBackProp (VBP) [11].”
Textual Explanations for Self-Driving Vehicles (2018, background): “[2] develop a richer notion of contribution of a pixel to the output.”
Deep Steering: Learning End-to-End Driving Model from Spatial and Temporal Visual Cues (2017, methodology): “DAVE-2 [7], [11] or PilotNet [23] were inspired by ALVINN and DAVE.”
Transparency in Deep Reinforcement Learning Networks (2018, methodology): “Other notable sensitivity approaches include a layer wise pixel decomposition method [1], visual backpropagation [4], guided backpropagation”
Interpretable Learning for Self-Driving Cars by Visualizing Causal Attention (2017, background): “[2] developed a richer notion of ”contribution” of a pixel to the output.”
Visual Analytics in Deep Learning: An Interrogative Survey for the Next Frontiers (2018, unknown): “One technique called visual backpropagation attempts to visualize what parts of an image contributed to the classification, but does so in real-time in a model debugging tool for self-driving vehicles [30].”
A Survey of Deep Learning for Scientific Discovery (2020, background): “Many other techniques look at computing some kind of input mask, several of them using deconvolutional layers, ﬁrst proposed by [263] and built on by [106] and [20].”
Deep-learning in Mobile Robotics - from Perception to Control Systems: A Survey on Why and Why not (2016, methodology): “DeepDriving (Chen et al., 2015) estimated several indicators for affordance of driving actions using CNN models, but not directly mapping sensor input to driving commands (Muller et al., 2005; Giusti et al., 2016; Tai et al., 2016; Bojarski et al., 2016).”