RealEstate10K
PulseAugur coverage of RealEstate10K — every cluster mentioning RealEstate10K across labs, papers, and developer communities, ranked by signal.
12 day(s) with sentiment data
-
GibbsLens: New Training-Free Camera Control for Video Generation
Researchers have developed GibbsLens, a novel training-free method for camera control in video generation. This approach enhances existing flow-matching video generators by refining latent sites through a block-conditio…
-
New methods enhance 3D Gaussian Splatting for reconstruction and synthesis · 6 sources tracked
Researchers have developed several new methods to enhance 3D Gaussian Splatting (3DGS), a technique for novel view synthesis and 3D reconstruction. OctreeSplat introduces a hierarchical octree structure guided by visual…
-
SG-Splat improves 3D Gaussian Splatting robustness to pose uncertainty
Researchers have introduced SG-Splat, a novel framework designed to enhance the robustness of pose-free 3D Gaussian Splatting (3DGS) against pose uncertainty. This method separates geometry estimation from Gaussian attr…
-
FPGA-based LENS system brings GPU-level scene memory to edge robots
Researchers have developed LENS, a system designed to enable resource-constrained robots to perform scene memory tasks without relying on high-power desktop GPUs. This system utilizes an FPGA and ARM platform to acceler…
-
Kyutai unveils CADRE video autoencoders for faster interactive world models
Kyutai has introduced CADRE, a new set of video autoencoders designed to reduce latency in interactive world models. These autoencoders aim to improve upon existing models by processing video frames more efficiently, ac…
-
New framework CausalAlign enhances video generation model ego-motion consistency
Researchers have developed CausalAlign, a new framework designed to improve the consistency of camera ego-motion in diffusion-based video generation models (VGMs). The framework uses causal interventions to identify how…
-
New CausalAlign framework boosts video generation model ego-motion consistency
Researchers have developed CausalAlign, a new framework designed to improve the consistency of camera ego-motion in diffusion-based video generation models. The framework addresses the issue of ego-motion drift over lon…
-
HRoPE introduces surface-aware positional encoding for multi-view transformers
Researchers have introduced HRoPE (Homography Relative Positional Encoding), a novel method for improving spatial understanding in multi-view transformers. Unlike previous approaches that focus on camera geometry, HRoPE…
-
LonGaussian framework enables 3D scene reconstruction from long videos
Researchers have introduced LonGaussian, a novel feed-forward framework designed for reconstructing high-quality 3D scenes from long video sequences. This method addresses memory and storage limitations by processing vi…
-
New CoTL framework learns compact 3D scene tokens from multi-view images
Researchers have introduced CoTL, a novel framework designed for learning compact token representations of 3D scenes from multi-view images. This method compresses input images into a small set of latent tokens that agg…
-
GeoBridge framework evolves transferable decoder topologies for geometric vision
Researchers have introduced GeoBridge, an evolutionary framework designed to discover transferable decoder topologies for geometric vision tasks. This system optimizes internal computations within decoders, allowing the…
-
CompactSplat improves 3D scene reconstruction with adaptive Gaussian allocation
Researchers have introduced CompactSplat, a new feedforward framework designed to improve 3D scene reconstruction from sparse views. This method adaptively allocates 3D Gaussians, concentrating them in complex areas whi…
-
Zebra3D framework boosts 3D reconstruction efficiency with hybrid attention
Researchers have introduced Zebra3D, a novel framework designed to enhance the efficiency of 3D reconstruction models. This hybrid attention approach selectively applies linear attention to less sensitive layers and spa…
-
New models tackle novel view synthesis and camera pose estimation
Two new research papers introduce novel methods for synthesizing new views of a scene from existing images and estimating camera poses. The first paper, LVS, focuses on reusing previously rendered images to improve inte…
-
CompactSplat framework improves 3D scene reconstruction with adaptive Gaussian allocation
Researchers have introduced CompactSplat, a new feedforward framework designed for reconstructing 3D scenes from sparse views. This method adaptively allocates 3D Gaussians using texture-guided spatial partitioning and …
-
New frameworks boost 3D novel view synthesis efficiency and quality · 3 sources tracked
Researchers have developed three new frameworks for novel view synthesis, aiming to improve efficiency and quality in 3D content creation. NAMVIS reformulates the process as autoregressive next-scale prediction, outperf…
-
PoseAgent framework dynamically orchestrates camera pose estimation methods
Researchers have developed PoseAgent, a novel framework designed to improve the accuracy and robustness of relative camera pose estimation. This system dynamically selects and orchestrates various pose estimation method…
-
Honeycomb video model uses fixed-size HexMemory for consistent scene representation
Researchers have introduced Honeycomb, a novel video world model that utilizes a fixed-size scene memory called HexMemory. This HexMemory representation employs a low-rank factorization across six spatial and spatiotemp…
-
New SPAR3S model generates 3D scenes from sparse multi-view images
Researchers have developed SPAR3S, a novel sparse auto-regressive model for generating complete 3D scenes from limited multi-view images. This method operates in a compact, voxel-aligned 3D latent space, representing on…
-
New methods enhance novel view synthesis stability and quality · 2 sources tracked
Researchers have developed new methods to improve the stability and quality of novel view synthesis, a technique that generates new perspectives of a scene from a limited set of input images. One approach, CamTrol++, fo…