test the deep learning for rigid registration
the folder of data contains 100 images (about 5 M)
Run: python main.py
I use the CNN to predict the rotation angle. However, the loss is very large:
iteration: 1476 || loss = 0.102 || error (degree) = 49.990
iteration: 1477 || loss = 0.077 || error (degree) = 42.407
iteration: 1478 || loss = 0.075 || error (degree) = 45.109
iteration: 1479 || loss = 0.015 || error (degree) = 17.671
iteration: 1480 || loss = 0.039 || error (degree) = 31.031
iteration: 1481 || loss = 0.034 || error (degree) = 25.559
iteration: 1482 || loss = 0.086 || error (degree) = 48.926
iteration: 1483 || loss = 0.043 || error (degree) = 32.719
iteration: 1484 || loss = 0.051 || error (degree) = 32.231
iteration: 1485 || loss = 0.101 || error (degree) = 48.668
iteration: 1486 || loss = 0.043 || error (degree) = 31.668
The error is: abs(pred_rotation_angle - true_rotation_angle). The error is very large.