Neural Point-Based Graphics

Free View Synthesis

“…The results are summarized in Table 1a. The first row reports the accuracy with the Weighted Mean aggregation as described in Equation (2). The second and third rows report accuracy with the MLP aggregation function (see Equation (3)), once with the mean and once with the max pooling operator.…”

“…For Extreme View Synthesis (EVS) [6] we also use the publicly available code and the provided network weights. Neural Point Based Graphics (NPBG) [2] is fitted per scene using the published code NeRF++ [43] NPBG [2] M60 Playground Train Figure 5: Qualitative results on Tanks and Temples. Comparison of SVS to the best-performing prior methods.…”

“…FVS [29] NeRF++ [43] NPBG [2] Bike Sandbox Soccertable Figure 6: Qualitative results on the FVS dataset. Comparison of SVS to the best-performing prior methods.…”

“…NPBG [2] is based on a reconstructed point cloud of the scene. As outlined above, this can be computed in <65 minutes.…”

SCALE: Modeling Clothed Humans with a Surface Codec of Articulated Local Elements

“…As the aforementioned meshing methods are sensitive to hyperparameters, we present a method to directly render the SCALE outputs into high-resolution images by leveraging neural rendering based on point clouds [1,57,81]. In Sec.…”