-
Ghost-DeblurGAN
This is a lightweight GAN developed for real-time deblurring. The model has a super tiny size and a rapid inference time. The motivation is to boost marker detection in robotic applications, however, you may use it for other applications definitely.
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
You could try some sort of Image restoration GAN to produce a modified output of the input image (which lies on the curved surface). I have conducted research on GANs for deblurring Fiducial markers, and simply put, the Network takes a blurred image containing fiducial markers and outputs a ‘deblurred image’ whose fiducial markers can be more easily detected. However, keep in mind that the GANs can perform a wide variety of operations on an input Image (other than just deblurring), hence, they are called Image restoration networks in general. What you could give a shot is, create a pipeline where the captured image goes into the GAN, then the output is fed to the Apriltag detector. Here is the link to our GitHub if you need a starter: https://github.com/York-SDCNLab/Ghost-DeblurGAN, and some of my other suggestions are NAFNet and HINet.