There has been a rise in the use of Machine Learning as a Service (MLaaS)Vision APIs as they offer multiple services including pre-built models and algorithms . As these APIs get deployed for high-stakes applications, it’s veryimportant that they are robust to different manipulations . We propose two new aspects of adversarial image generation methods and evaluate them on the robustness of Google Cloud Vision API’s opticalcharacter recognition service and object detection APIs deployed in real-world settings . Our transparent adversarial examples successfullyevade state-of-the art object detections APIs such as . Azure Cloud Vision(attack success rate 52%) and Google Cloud .com, picpurify.com, Google Cloud vision API, andMicrosoft Azure’s Computer Vision API, respectively, were found to have a successful attack success rate of 36% of the images with a secret embedded text that successfully fools the time-limited humans but is detected by Google CloudVision API’soptical character recognition . As a result, they pose a serious threat where APIs are used for time- limited humans but are detected by

Author(s) : Jaydeep Borkar, Pin-Yu Chen

Links : PDF - Abstract

Code :
Coursera

Keywords : vision - apis - api - cloud - google -

Leave a Reply

Your email address will not be published. Required fields are marked *