RIT-NTNU-Bachelor / case-study

Repository for comparing face detection algorithms, focusing on memory usage and accuracy. Includes benchmarks and evaluations to guide optimal algorithm selection.

MIT License

1 stars 0 forks source link

Case study: distances estimation (standarized test) #5

Closed KjetilIN closed 4 months ago

KjetilIN commented 5 months ago

What?

Add tests for the distance estimation method

KjetilIN commented 5 months ago

Think the best solution is to take two images from this dataset: https://robotics.dei.unipd.it/reid/index.php/8-dataset/9-overview-face

KjetilIN commented 5 months ago

Dataset:

IAS-Lab RGB-D Face dataset:

Using the provided

Link to the dataset home page: https://robotics.dei.unipd.it/reid/index.php/8-dataset/9-overview-face

Link to the dataset download page: https://robotics.dei.unipd.it/reid/index.php/downloads

Citations

Authors of dataset and paper asks to use citation if we use the dataset

G. Pitteri, M. Munaro and E. Menegatti. "Depth-based frontal view generation for pose invariant face recognition with consumer RGB-D sensors". In Proceedings of the 14th International Conference on Intelligent Autonomous Systems (IAS-14), Shanghai, 2016.

KjetilIN commented 5 months ago

Needs review

Struggle to test. Gets as a result that the distance is 216 cm for images where it should be 100 cm. Dont know how to fix. Dont know what is wrong. No clear way to find focal lenght or the sensor width of the camera. No realiable way of testing. Either the equation is wrong or the units in the equation does not correlate

@marhjoh @haugeSander

marhjoh commented 5 months ago

@KjetilIN I would assume it is the code, and not your tests. After some research I located a similar issue (regarding hand tracking) with CVZone. Might need some thorough refactoring, but worth to take a look at that.

MediaPipe Issue, and more specifically this answer.

Let me know if you want me to help you with this.

KjetilIN commented 4 months ago

Up for review

Check PR #8