3D Gaussian Splatting (3DGS) has recently emerged as a promising contender to Neural Radiance Fields (NeRF) in 3D scene reconstruction and real-time novel view synthesis. 3DGS outperforms NeRF in training and inference speed but has substantially higher storage requirements. To remedy this downside, we propose POTR, a post-training 3DGS codec built on two novel techniques. First, POTR introduces a novel pruning approach that uses a modified 3DGS rasterizer to efficiently calculate every splat’s individual removal effect simultaneously. This technique results in 2-4× fewer splats than other post-training pruning techniques and as a result also significantly accelerates inference with experiments demonstrating 1.5-2× faster inference than other compressed models. Second, we propose a novel method to recompute lighting coefficients, significantly reducing their entropy without using any form of training. Our fast and highly parallel approach especially increases AC lighting coefficient sparsity, with experiments demonstrating increases from 70% to 97%, with minimal loss in quality. Finally, we extend POTR with a simple fine-tuning scheme to further enhance pruning, inference, and rate-distortion performance. Experiments demonstrate that POTR, even without fine-tuning, consistently outperforms all other post-training compression techniques in both rate-distortion performance and inference speed.

Sending low-delay live video over error-prone channels comes with packet-loss-repair and random-access challenges. Existing solutions have a negative impact on end-users with reliable connections or users that do not switch channels. To minimize this impact, the keyframe-insertion technique extends a compression-efficient normal stream (NS) with a companion stream (CS) solely consisting of keyframes [1].

In six-degrees-of-freedom light-field (LF) experiences, the viewer's freedom is limited by the extent to which the plenoptic function was sampled. Existing LF datasets represent only small portions of the plenoptic function, such that they either cover a small volume, or they have limited field of view. Therefore, we propose a new LF image dataset "SILVR" that allows for six-degrees-of-freedom navigation in much larger volumes while maintaining full panoramic field of view. We rendered three different virtual scenes in various configurations, where the number of views ranges from 642 to 2226. One of these scenes (called Zen Garden) is a novel scene, and is made publicly available. We chose to position the virtual cameras closely together in large cuboid and spherical organisations (2.2m3 to 48m3), equipped with 180° fish-eye lenses. Every view is rendered to a color image and depth map of 2048px × 2048px. Additionally, we present the software used to automate the multiview rendering process, as well as a lens-reprojection tool that converts between images with panoramic or fish-eye projection to a standard rectilinear (i.e., perspective) projection. Finally, we demonstrate how the proposed dataset and software can be used to evaluate LF coding/rendering techniques (in this case for training NeRFs with instant-ngp). As such, we provide the first publicly-available LF dataset for large volumes of light with full panoramic field of view.

Forensic watermarking is used to track down digital pirates after they illegally redistribute video content. Although existing algorithms often resist common signal processing attacks, they are not always robust against camcording attacks. As a solution in the state of the art, registration methods are used to align the attacked video to the original one. However, watermark detection still fails when the quality is sufficiently decreased or when exposed to targeted attacks. Therefore, this paper proposes a novel fallback system that aims to detect the watermark when traditional methods fail. More concretely, we demonstrate that a primary watermark embedded by a traditional scheme indirectly creates a secondary watermark signal during video encoding. This secondary watermark consists of compression artifacts and is detected by the fallback system. Additionally, the proposed system incorporates video registration to cope with camcording attacks. The experimental results indicate that the fallback system has a striking increase in robustness compared to the existing methods. For example, the observed false-negative rate for targeted attacks improves from 100% to 0%. Moreover, the fallback is camcording resistant even when the traditional method combined with registration is not. In conclusion, the proposed system can be used as a fallback when traditional detection fails.

Steered Mixture-of-Experts (SMoE) is a novel framework for the approximation, coding, and description of image modalities. The future goal is to arrive at a representation for Six Degrees-of-Freedom (6DoF) image data. The goal of this paper is to introduce SMoE for 4D light field videos by including the temporal dimension. However, these videos contain vast amounts of samples due to the large number of views per frame. Previous work on static light field images mitigated the problem by hard subdividing the modeling problem. However, such a hard subdivision introduces visually disturbing block artifacts on moving objects in dynamic image data. We propose a novel modeling method that does not result in block artifacts while minimizing the computational complexity and which allows for a varying spread of kernels in the spatio-temporal domain. Experiments validate that we can progressively model light field videos with increasing objective quality up to 0.97 SSIM.