DataStore Length Mismatch b/w image and bbox

2 visualizaciones (últimos 30 días)
Kev el 29 de Oct. de 2020
Comentada: Kev el 2 de Nov. de 2020
I am trying to run YOLO example with my own dataset. I am currently attempting with a small batch of files (181). I have 181 image files and 200 bbox files.
Can you combine and run a test when the length is different? Some images have been removed on purpose.
I am getting the error on this line,
detectionResults = detect(detector, preprocessedTestData);
Also,do you have to have a bbox for a particular imagefile, can it skip the file if there is no bbox?

Respuesta aceptada

Divya Gaddipati
Divya Gaddipati el 2 de Nov. de 2020
Each image should contain atleast one labeled object and ideally more labeled images are required to train a robust network.
Since, you mentioned that some of the images have been removed, their corresponding bounding boxes should also be removed.
If any image doesn't have any bbox, that sample should be removed as well.
Assuming you are using R2020b, you can refer to the validateInputData attached to the YOLOv3 example. This function removes the invalid samples as mentioned in the example.
  1 comentario
Kev el 2 de Nov. de 2020
Thanks Divya, I was able to resolve that error.
I have two other questions, hopefully you can asnwer them as well
1) Does MATLAB have the official YOLO model/weights that one can use? The model in above example seems to have been trained on those particular images only. I have tied using ONXX route, but I get error. I opened another question for it, but waiting for response right now.
2) Can one limit what classes they want to detect rather than idetifying everyhing? If you're interested in partcular classes, can we pre-define that?
Thanks a lot!!

Iniciar sesión para comentar.

Más respuestas (0)


Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by