Neural Pixel Composition for 3D-4D View Synthesis from Multi-Views

Figure 1 . Our approach allows us to capture small details better than existing methods. We show novel views (top-row) synthesized using our approach and zoom on the details for each view (bottom-row). Our model is trained for 10 minutes. Best viewed in electronic format.