# Towards Safety Verification of Direct Perception Neural Networks

arXiv: https://arxiv.org/abs/1904.04706 · HTML: https://mlsanity.com/paper/1904.04706/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 9 Apr 2019.

## Built on by (most cited first, 9 of 9)

- [Runtime Monitoring for Out-of-Distribution Detection in Object Detection Neural Networks](https://mlsanity.com/paper/2212.07773/) (2022)
- [BAM: Box Abstraction Monitors for Real-time OoD Detection in Object Detection](https://mlsanity.com/paper/2403.18373/) (2024)
- [Certifying Robustness of Learning-Based Keypoint Detection and Pose Estimation Methods](https://mlsanity.com/paper/2408.00117/) (2024)
- [Prioritizing Corners in OoD Detectors via Symbolic String Manipulation](https://mlsanity.com/paper/2205.07736/) (2022)
- [Backdoor Mitigation in Deep Neural Networks via Strategic Retraining](https://mlsanity.com/paper/2212.07278/) (2022)
- [Runtime Monitoring DNN-Based Perception](https://mlsanity.com/paper/2310.03999/) (2023)
- [Safe AI -- How is this Possible?](https://mlsanity.com/paper/2201.10436/) (2022)
- [Attack logics, not outputs: Towards efficient robustification of deep neural networks by falsifying concept-based properties](https://mlsanity.com/paper/2510.03320/) (2025)
- [From Decoupled to Coupled: Robustness Verification for Learning-based Keypoint Detection with Joint Specifications](https://mlsanity.com/paper/2603.05604/) (2026)

## Abstract (arXiv)

We study the problem of safety verification of direct perception neural networks, where camera images are used as inputs to produce high-level features for autonomous vehicles to make control decisions. Formal verification of direct perception neural networks is extremely challenging, as it is difficult to formulate the specification that requires characterizing input as constraints, while the number of neurons in such a network can reach millions. We approach the specification problem by learning an input property characterizer which carefully extends a direct perception neural network at close-to-output layers, and address the scalability problem by a novel assume-guarantee based verification approach. The presented workflow is used to understand a direct perception neural network (developed by Audi) which computes the next waypoint and orientation for autonomous vehicles to follow.