Benchmarking a rebuilt decoder honestly
I timed the rebuilt VCode decoder against the old one on the same phone. Same speed per frame; the gains were size, reach and compatibility.
What I tried
After rebuilding the decoder I wanted a straight answer on performance, not an impression from using it.
Why I chose differently
Comparing two builds on feel tells you nothing. They have to do the same job on the same phone with the same input.
What I chose
I ran the old and new decoders side by side on the same Pixel with identical frames. They run at the same speed, because the core work is the same. The gains are elsewhere: a quarter of the size, at home on current Android and able to read codes anywhere in the picture.
The principle
Report what you measured, including when it is not the headline you hoped for. "Same speed, a quarter of the size, works on new phones" is a better result to stand behind than a speed claim that would not survive a test.
- Benchmarking
- Performance
- Android NDK
- Engineering honesty