Improve performance and optimize memory

[rainyl]

Reduce the consumption, improve performance

• Optimize the use of Arenas
• Add dispose() to all classes with native pointer
• Add detach to the constructors with native pointer

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 89.98779% with 82 lines in your changes are missing coverage. Please review.

Project coverage is 89.50%. Comparing base (a04f909) to head (a6e0907).

Files	Patch %	Lines
lib/src/core/point.dart	80.30%	13 Missing ⚠️
lib/src/imgproc/imgproc.dart	81.81%	10 Missing ⚠️
lib/src/core/cv_vec.dart	94.80%	8 Missing ⚠️
lib/src/core/mat.dart	79.48%	8 Missing ⚠️
lib/src/core/rect.dart	73.33%	8 Missing ⚠️
lib/src/highgui/highgui.dart	27.27%	8 Missing ⚠️
lib/src/core/array.dart	80.00%	7 Missing ⚠️
lib/src/core/asyncarray.dart	28.57%	5 Missing ⚠️
lib/src/dnn/dnn.dart	83.87%	5 Missing ⚠️
lib/src/features2d/features2d.dart	96.69%	4 Missing ⚠️
... and 3 more

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #67      +/-   ##
==========================================
+ Coverage   88.97%   89.50%   +0.53%     
==========================================
  Files          36       36              
  Lines        4897     5165     +268     
==========================================
+ Hits         4357     4623     +266     
- Misses        540      542       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rainyl]

Ready to merge.

Note:

• Wrapper classes with large native resources, users now can free them manually with dispose() to free native resources, but note that the dart instances will still be available, users have to ensure not to access them if disposed, dart instances are small and believe dart VM/GC to manage them.

Example:

import 'package:opencv_dart/opencv_dart.dart' as cv;

void main() {
  const counts = 10000;
  final mat = cv.Mat.zeros(1000, 1000, cv.MatType.CV_8UC3);
  final sw = Stopwatch()..start();
  for (var count = 0; count < counts; count++) {
    final mat1 = cv.cvtColor(mat, cv.COLOR_BGR2YCrCb);
    // manually dispose will reduce the memory consumption
    mat1.dispose();
  }
  sw.stop();
  print(
      "All: ${sw.elapsedMicroseconds}μs, counts: $counts, per: ${sw.elapsedMicroseconds / counts} μs");
}

Without mat1.dispose(), peak memory: ~18G

All: 5527358μs, counts: 10000, per: 552.7358 μs

With mat1.dispose(), peak memory: ~240M

All: 7263633μs, counts: 10000, per: 726.3633 μs

[abdelaziz-mahdy]

Ready to merge.

Note:

• Wrapper classes with large native resources, users now can free them manually with dispose() to free native resources, but note that the dart instances will still be available, users have to ensure not to access them if disposed, dart instances are small and believe dart VM/GC to manage them.

Example:

import 'package:opencv_dart/opencv_dart.dart' as cv;

void main() {
  const counts = 10000;
  final mat = cv.Mat.zeros(1000, 1000, cv.MatType.CV_8UC3);
  final sw = Stopwatch()..start();
  for (var count = 0; count < counts; count++) {
    final mat1 = cv.cvtColor(mat, cv.COLOR_BGR2YCrCb);
    // manually dispose will reduce the memory consumption
    mat1.dispose();
  }
  sw.stop();
  print(
      "All: ${sw.elapsedMicroseconds}μs, counts: $counts, per: ${sw.elapsedMicroseconds / counts} μs");
}

Without mat1.dispose(), peak memory: ~18G

All: 5527358μs, counts: 10000, per: 552.7358 μs

With mat1.dispose(), peak memory: ~240M

All: 7263633μs, counts: 10000, per: 726.3633 μs

Does this mean the package user will have to manage the object? I another way do I have to dispose every native asset I use?

[rainyl]

package user will have to manage the object?

Ready to merge.
Note:

• Wrapper classes with large native resources, users now can free them manually with dispose() to free native resources, but note that the dart instances will still be available, users have to ensure not to access them if disposed, dart instances are small and believe dart VM/GC to manage them.

Example:

import 'package:opencv_dart/opencv_dart.dart' as cv;

void main() {
  const counts = 10000;
  final mat = cv.Mat.zeros(1000, 1000, cv.MatType.CV_8UC3);
  final sw = Stopwatch()..start();
  for (var count = 0; count < counts; count++) {
    final mat1 = cv.cvtColor(mat, cv.COLOR_BGR2YCrCb);
    // manually dispose will reduce the memory consumption
    mat1.dispose();
  }
  sw.stop();
  print(
      "All: ${sw.elapsedMicroseconds}μs, counts: $counts, per: ${sw.elapsedMicroseconds / counts} μs");
}

Without mat1.dispose(), peak memory: ~18G

All: 5527358μs, counts: 10000, per: 552.7358 μs

With mat1.dispose(), peak memory: ~240M

All: 7263633μs, counts: 10000, per: 726.3633 μs

Does this mean the package user will have to manage the object? I another way do I have to dispose every native asset I use?

Not necessary, the above test is for dart, I didn't test in Flutter but Flutter will trigger GC more frequently than pure dart, so it tends to rely on GC to free, BUT, if users can know that the APP will process large images (e.g., 3840*2160*3), it's recommend to dispose manually.

Anyway, this PR just provides a potential to manage native resources more efficiently according to the weakness of the GC of dart VM, but not necessary.

[abdelaziz-mahdy]

Oh got it , so GC works too but if needed manual mode can be used for optimization

[rainyl]

Oh got it , so GC works too but if needed manual mode can be used for optimization

bingo!