train dataset of PointHistoryClassifier? about hand-gesture-recognition-using-mediapipe

[brooks0519]

I am so sorry to disturb you, I have find the reason of the prroblem. because after I train my dataset in the notebook, I didn't run the part h5->tflite, so the inference model is also the original model you give. After I run the h5->tflite, the demo app.py can get the right result as expect. Thanks again for your great job!

[Kazuhito00]

Please check and tell me some things to solve the problem.

• Did you change the NUM_CLASSES in "point_history_classification.ipynb"?
• Did you add a new class label to "point_history_classifier_label.csv"?
• Check the confusion matrix and find a lot of wrong result?

If all of the above are okay, share your train data and notebook.

[brooks0519]

Thank you for your replay.
firstly, I have change the NUM_CLASSES in "point_history_classification.ipynb" to 5(as I add a new gesture);
secondly, I have add a new class label to "point_history_classifier_label.csv" which index is 4;
thirdly, the matrix is as bellow:

and the train data and notebook is here:
point_history_classification.zip

my new gesture is like this in the below pic:

[Kazuhito00]

I'm glad I solved it.