I am creating an Action Detection Sign Language Model but no matter which dataset I use I always get in the problem of having videos with different numbers of frames which causes an error when structuring the data or setting the train_test_split. I looked for alternative options of datasets but I was unlucky to find any. This is a college project and I just need to get a DEMO running. I used this method that I got from GPT: max_len = max(len(seq) for seq in sequences) X = pad_sequences( sequences, maxlen = max_len, dtype = 'float32', padding = 'post', truncating = 'post') y = to_categorical(labels).astype(int) X_train, X_test, y_train, y_test = train_test_split( X, y, test_size = 0.05, random_state = 42, shuffle = True) but even with this method the model was not learning at all. it was picking randomly (between 3 classes the accuracy was 0.3333) Continue reading...
