Fig. 1.

Artifact disentanglement flowchart of PAT-ADN. The artifact-affected image $u^a$ obtained under sparse view sampling is encoded into both artifact space A and content space C. The artifact-free image $v$ obtained under full-view sampling is encoded into content space C. Encoding $u^a$ and $v$ into the content space C to generate the content encoding $c_u$ and $c_v$. Encoding $u^a$ into artifact space A to get artifact code $a$. By decoding the content code $c_u$, artifacts can be removed from the artifact-affected image (blue arrows $u^a\to u$). By decoding the artifact code $a$ and content code $c_v$, artifacts are introduced into artifact-free images, producing the image $v^a$ with artifacts (green arrows $v\to v^a$).