During training, I use videos consisting of 16,000 frames, processing them in batches of 200 frames each for the model. For each dataset, I achieve values close to the actual data at the end of training. However, when I start training with a different dataset, the initial predictions tend to be very close to the last predicted values of the previous dataset. In other words, instead of starting with an estimate based on the new dataset, it seems to predict values close to the last obtained results, seemingly independent of the input. I’m not sure if this indicates overfitting or if there’s another problem with my model. Does anyone have any thoughts on this?