Learning Iterative Image Reconstruction - Hierarchical Neural Networks for Image Interpretation - page 195

8

7 TRN high noise

TST high noise
TRN medium noise
TST medium noise
TRN low noise
TST low noise

6

5

4

3

2

1

0

5 10 15 20

iteration

Fig. 9.28. Output changes over time for the reconstruction from a sequence of degraded
MNIST digits. Performance for the training set (TRN) and the test set (TST) is very similar.
The outputs change most during the first iterations. Output changes are higher for high noise
than for lower noise. The bump at iteration 17 reflects a sudden change in the input sequence
that starts again with the first frame.

(a) 1 2 4 7 11 16 (b) 1 2 4 7 11 16 (c)
Fig. 9.29. Reconstruction from a sequence of degraded MNIST images (high noise): (a) input
over time; (b) output over time; (c) target.