Fig. 5.

Comparisons of the reconstruction performance with and without the progressive training-in-time strategy using d = 40. From the plot of SER vs. running time, we can see that the progressive training-in-time approach yields better results with much less running time comparing to the training without using progressive training-in-time. Two reconstructed frames near the end of systole and diastole using SToRM500, the proposed scheme with progressive training-in-time and the proposed scheme without using the progressive training-in-time are shown in the plot as well for comparison purposes. The average Brisque scores for SToRM500, the reconstruction with progressive training-in-time, and the reconstruction without progressive training-in-time are 36.4, 37.3 and 39.1 respectively.