Hello everyone,
I am currently writing my bachelor’s thesis on computer vision and deep learning. My topic is an industrial screwdriving process that, in a few cases, causes screws to fall out of the nozzle in an uncontrolled manner, which should be detected using deep learning. The process is filmed by a high-speed camera and the individual frames can be extracted (I have attached an example image). I have already experimented with Pytorch and the EfficientNet-B0 model (2 output classes in the FC layer → good, not good) and came across the Intel framework anomalib. I would like to receive feedback on which approach makes sense to pursue in my use case and what is the best approach for detecting this critical process.
→ The frame rate of the not good image is pretty bad in this example. The not good images in my dataset will be much clearer to identify than in this example.
