Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

CVT-xRF: Contrastive In-Voxel Transformer for 3D Consistent Radiance Fields from Sparse Inputs

About

Neural Radiance Fields (NeRF) have shown impressive capabilities for photorealistic novel view synthesis when trained on dense inputs. However, when trained on sparse inputs, NeRF typically encounters issues of incorrect density or color predictions, mainly due to insufficient coverage of the scene causing partial and sparse supervision, thus leading to significant performance degradation. While existing works mainly consider ray-level consistency to construct 2D learning regularization based on rendered color, depth, or semantics on image planes, in this paper we propose a novel approach that models 3D spatial field consistency to improve NeRF's performance with sparse inputs. Specifically, we first adopt a voxel-based ray sampling strategy to ensure that the sampled rays intersect with a certain voxel in 3D space. We then randomly sample additional points within the voxel and apply a Transformer to infer the properties of other points on each ray, which are then incorporated into the volume rendering. By backpropagating through the rendering loss, we enhance the consistency among neighboring points. Additionally, we propose to use a contrastive loss on the encoder output of the Transformer to further improve consistency within each voxel. Experiments demonstrate that our method yields significant improvement over different radiance fields in the sparse inputs setting, and achieves comparable performance with current works.

Yingji Zhong, Lanqing Hong, Zhenguo Li, Dan Xu• 2024

Related benchmarks

TaskDatasetResultRank
Novel View SynthesisDTU 6-view
PSNR25.5
49
Novel View SynthesisDTU Object 3-view
PSNR21.51
14
Novel View SynthesisDTU Object 9-view
PSNR27.68
14
Novel View SynthesisDTU Full-image 3-view
PSNR18.98
13
Novel View SynthesisDTU Full-image 9-view
PSNR27.04
13
Novel View SynthesisDTU 3-view
LPIPS (Full-image)0.187
12
Novel View SynthesisDTU 9-view
Full-image LPIPS0.074
12
Novel View SynthesisSynthetic 3-view
PSNR21.58
11
Novel View SynthesisSynthetic 3-view (test)
LPIPS0.176
11
Novel View SynthesisSynthetic 8-view
LPIPS0.108
9
Showing 10 of 10 rows

Other info

Code

Follow for update