Hello,
I am currently working on a project involving the classification of a judo competition. I have trained a tenerflow.js model to detect Ipon, Wazari and Yuko in live footage. I then implemented the model into a simple web application that runs locally on my systems. The app works fine, but no matter what video I show the app, it always predicts the first category of this array: const CLASS_NAMES = [‘Ipon’, ‘Wazari’, ‘Yuko’] so in this case Ipon, even if the video does not show an Ipon. All three classes are trained with the same amount of data. The probability for this example is
Ipon: 84.8% Wazari: 6.1% Yuko: 9.1% even though the Viedio does not show an Ipon. All three classes are trained with 40 videos, which will be expanded in the future as I know the model works.
Thanks in advance for any tips or advice on how to fix this problem, if you have any questions please do not hesitate to contact me,
kind regards
Peter