Isaac data quality

[MinhxNguyen7]

Description

From the dataset of 7631 images, there are 37 classes generated:

['background', 'l', 'p', 'u', 'triangle', 'quarter_circle', 'cross', 'w', 'n', 'pentagon', 'star', 'd', 'j', 'semi_circle', 'g', 'circle', 'z', 'rectangle', 't', 'v', 'a', 'h', 'c', 'y', 'm', '7', 'k', 'e', 'f', 'r', 'q', 'x', '8', 'i', 's', '6', 'o']

Notably, that list is missing a letter (b), and quite a few numbers. At least it includes all the shapes. Not sure if this is missing from the Isaac side or if there's a mistake in translating the data.

References

• Dataset requirements for training: https://docs.ultralytics.com/yolov5/tutorials/tips_for_best_training_results/#:~:text=YOLOv5%20trainings%20below.-,Dataset,-Images%20per%20class

[MinhxNguyen7]

@Dat-Bois

[Dat-Bois]

Pretty sure it is Isaac side. Quickest way to verify would be to write a small script that just goes through all of the isaacs label files, parses the class info, and adds it to a set. Print the set at the end and you should have all the genned classes from isaac.