Training and Testing is looping over a similar range of frames

[blank-ed]

In pulse_data_3d.py, when printing the frame index ranges for each batch—determined by idx and idx+self.seq_len—there's an evident pattern. Initially, the iterations cover up to index 4010 before reverting to index 2151. As training progresses, this maximum index incrementally increases by several hundred. Essentially, the model retrains on the same frames. When using the whole PURE dataset, the training appears confined to the initial few videos, specifically 01-01, 01-02, and 01-03.