nstrndrbi / starnet Goto Github PK

Nastaran Darabi1*   Sina Tayebati1*   Sureshkumar S.¹   Sathya Ravi¹   Theja Tulabandhula¹   Amit R. Trivedi^1
1University of Illinois Chicago    ^*Both authors contributed equally to this work   

STARNet introduces a deep network designed for anomaly/corruption detection in LiDAR-Camera data pipelines, with focus on implementation on edge AI devices such as micro drones. STARNet employs a gradient-free likelihood regret concept integrated with a variational autoencoder, making implementation on low-complexity edge hardware possible.

Bellow we demonstrate a demo of STARNet integrated with AIRSim, detecting OOD and IN-D data in online streaming.

We're cleaning the code for final release with full documentation