Human-Action-Recognition

I built a project for recognizing human actions using ConvLSTM and LRCN models, trained on a dataset from Kaggle with 50 action classes. Due to limited resources, I trained the model on only 5 classes. Dataset: https://www.kaggle.com/datasets/vineethakkinapalli/ucf50-action-recognition-dataset. I have just taken 5 classes from the whole dataset and in each class has 150+ videos on which 1) convLSTM giving me 100% training accuracy and validation accuracy 82% on 50 epochs but i have restore the waights of the model on minmumm loss. 2) I have also trained LRCN model on the given data LRCN training accuracy is 95% and validation 84% and test accuracy is 83%.