HyperThumbnail Super Resolution
As part of UROP Project, I worked with Prof. Qifeng Chen (陳启峰) (https://cqf.io/) on multiple Deep Learning Problems.
The only one worth mentioning (where progress was made) was HyperThumbail Super Resolution.
Suppose we have a large image stored somewhere that we wish to quickly prieview in lower resolution. However, we also wish to get back the larger imagine in reasonable time and accuracy.
The original idea is to introduce AI models within the JPEG Compression algorithm. I worked around with different architectures and ended up improving the accuracy of the previous model at a fraction of the speed tradeoff.