1: \begin{abstract}

2: 

3: %

4: We present Neural Pixel Composition (NPC), a novel approach for continuous 3D-4D view synthesis given only a discrete set of multi-view observations as input.

5: %

6: Existing state-of-the-art approaches require dense multi-view supervision and an extensive computational budget.

7: %

8: The proposed formulation reliably operates on sparse and wide-baseline multi-view imagery and can be trained efficiently within a few seconds to $10$ minutes for hi-res (12MP) content, i.e., 200-400$\times$ faster convergence than existing methods. 

9: %

10: Crucial to our approach are two core novelties:

11: %

12: 1) a representation of a pixel that contains color and depth information accumulated from multi-views for a particular location and time along a line of sight,

13: %

14: and 2) a multi-layer perceptron (MLP) that enables the composition of this rich information provided for a pixel location to obtain the final color output.

15: %

16: We experiment with a large variety of multi-view sequences, compare to existing approaches, and achieve better results in diverse and challenging settings.

17: %

18: Finally, our approach enables dense 3D reconstruction from sparse multi-views, where COLMAP, a state-of-the-art 3D reconstruction approach, struggles.

19: %

20: 

21: \keywords{3D View Synthesis, 4D Visualization, 3D Reconstruction.}

22: 

23: \end{abstract}

24: