Binary Function Similarity Detection (BFSD) is a foundational technique in software security, underpinning a wide range of applications including vulnerability detection, malware analysis. Recent advances in AI-based BFSD tools have led to significant performance improvements. However, existing evaluations of these tools suffer from three key limitations: a lack of in-depth analysis of performance-influencing factors, an absence of realistic application analysis, and reliance on small-scale or low-quality datasets.In this paper, we present the first large-scale empirical study of AI-based BFSD tools to address these gaps. We construct two high-quality and diverse datasets: BinAtlas, comprising 12,453 binaries and over 7 million functions for capability evaluation; and BinAres, containing 12,291 binaries and 54 real-world 1-day vulnerabilities for evaluating vulnerability detection performance in practical IoT firmware settings. Using these datasets, we evaluate nine representative BFSD tools, analyze the challenges and limitations of existing BFSD tools, and investigate the consistency among BFSD tools. We also propose an actionable strategy for combining BFSD tools to enhance overall performance (an improvement of 13.4%). Our study not only advances the practical adoption of BFSD tools but also provides valuable resources and insights to guide future research in scalable and automated binary similarity detection.

Determining how impactful an anomaly is on the viewing experience of an audience is important to production studios, content creators, and content distributors. However, judging the impact of each anomaly is a highly manual and subjective task. To support automation of this, we propose the use of noticeability and introduce a method of prediction. We employed a psychophysical experimental method to capture the impact of various anomalies across various images. We then developed a multi-scale context aggregation model, trained on that data, to predict the noticeability of anomalies on novel images. This notice ability prediction can then be used to prioritize anomalies and ensure that remediation efforts are spent where it would most benefit the audience experience.

Predicting image saliency has many potential useful applications across several industries, including film production, creative marketing content, product design, and quality control in manufacturing. Although image saliency prediction, in general, has made substantial progress with the advent of deep learning on benchmark datasets, it still has room to improve when applied to novel scenarios. This paper presents our experimental findings in dataset building and model development for predicting saliency of cinematic images in a real-world context. To evaluate the proposed model, we conducted experiments on both benchmark datasets (i.e., SALICON, MIT1003, MIT300, CAT2000) as well as a private dataset. The results demonstrate the superior performance of our proposed method. We highlight a film production use case, but the model and methods explored here may also generalize to other areas relevant to image saliency.