A while back I took a shot at a title, "Trek to Yomi", which contains ~260mb of H.264 video, encoded at an exceedingly silly bit-rate.
Linked here is a sample of my re-encoding.
The below figure should speak for itself, but after tuning ffmpeg a fair bit, I managed to get quite an extraordinary outcome, coupled with srep & lolz, a result of a 328x improvement with a near imperceivable loss in fidelity.
Has anyone else had any similar success stories? It's got me wondering how and why these algorithms were able to process these lower bit-rate frames in such an efficient manner compared to the original lossless files.
(And yes before anyone asks I did verify the archives to be valid)