Input only 2D DRR and predict the delta T from isocenter T in pose regression phase ?

[jiafenggang]

Multiple X-rays are collected during the DSA procedure. For example, in the DeepFluoro dataset, for each subject, there is one CT scan and between 24 - 111 X-rays. Each X-ray has its own isocenter matrix. If each original X-ray is input into the Encoder, the output delta T from the isocenter T will all be identity?
From the article and code implementation, i've learned that you assume the Isocenter position is a standard fixed one. In fact, it can change and will vary as the DSA robotic arm rotates around the patient.

I think the original CT volume and the 2D x-ray should be input to the Encoder together

I'm looking forward to your discussion. Thank you.