One of the main goals of this project was to be as energy efficient as possible.
For that purpose I did quite a bit of research and ended up with few contenders for both iOS and Android.
To test the contenders I used the provided sample image and each run was done 100 times, with the image always loading again from disk. This adds ~50ms of overhead to each time, but it's a good way to simulate a real world scenario.
The used benchmark can be found from the example project. For ios the benchmark was run on iPhone 7 and for Android using Pixel 3 emulator.
The main conteneders for iOS were:
ZXing had high promises for good performance but with quick benchmark it was quite clear that the native CIDetector was much faster.
| Library | avg time (ms) | mean time (ms) | image |
|---|---|---|---|
| CIDetector | 130.68 | 116 |  |
| zxing-cpp | 331.16 | 332 |  |
I ended up using CIDetector for iOS.
For Android the main contenders were:
BoofCV was the clear winner here. It was 1.8 times faster than the MLKit.
| Library | avg time (ms) | image |
|---|---|---|
| BoofCV | 101.05 |  |
| MLKit | 186.87 |  |
BoofCV also has the fabulous ability to down sample the image before processing it. This is a huge win for performance, if the qr codes are not too small.
By using inSampleSize of 2, the time was reduced to 80ms, but we could only read 2 of the 3 qr codes.