Objaverse
PulseAugur coverage of Objaverse — every cluster mentioning Objaverse across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
New WayPOP benchmark tests autonomous vehicle perception on unknown objects
Researchers have introduced WayPOP, a new benchmark designed to evaluate the perception systems of autonomous vehicles. This benchmark focuses on open-set panoptic tracking, meaning it can identify and track objects tha…
-
Transformer renders 3D Gaussian splats directly, bypassing rasterization
Researchers have developed GaussianRenderFormer, a novel transformer-based neural network capable of directly rendering 3D Gaussian splats into images without relying on traditional rasterization techniques. This approa…
-
New method enhances 3D MLLM recognition with functional reasoning
Researchers have developed a new method called PFIR (Part-level Functional Inference for Recognition) to improve the generalizability of 3D Multimodal Large Language Models (MLLMs). Current 3D MLLMs often rely on patter…
-
PhysLDM uses latent diffusion for high-fidelity deformable body simulation
Researchers have introduced PhysLDM, a novel framework for simulating deformable bodies with high fidelity. This system utilizes a spatiotemporal latent diffusion model, combined with a holistic variational auto-encoder…
-
New research identifies object-to-image ratio mismatch as key to visual retrieval failures
A new research paper identifies object-to-image (O2I) ratio mismatch as a primary cause of failure in visual instance retrieval, rather than resolution loss. The study, conducted on a benchmark of 3,021 Objaverse object…
-
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…
-
New methods accelerate Heat Kernel Textures computation
Researchers have developed two novel methods to accelerate the computation of Heat Kernel Textures (HKTex), a technique used for representing surface appearance. The first method, LocalHK, replaces spectral evaluation w…
-
UniPart introduces zero-shot language-grounded 3D part segmentation
Researchers have introduced UniPart, a novel feed-forward cross-modal 3D Transformer designed for zero-shot language-grounded 3D part segmentation. This model aims to overcome the limitations of existing 3D foundation m…
-
New MSVS-VAE model advances high-fidelity 3D reconstruction
Researchers have developed MSVS-VAE, a novel hierarchical set-based Variational Autoencoder designed for high-fidelity 3D reconstruction. This model addresses the limitations of existing methods by progressively densify…
-
DreamCAD framework generates editable 3D CAD models from unannotated meshes
Researchers have introduced DreamCAD, a novel multi-modal generative framework designed to create editable Boundary Representations (BReps) from 3D meshes without requiring CAD-specific annotations. The system utilizes …
-
New research introduces LATO.2 and Nexus for advanced 3D mesh generation · 4 sources tracked
Two new research papers introduce advanced methods for generating 3D meshes, a critical component for applications in film, gaming, and interactive 3D environments. LATO.2 employs a factorized flow matching framework th…
-
New 3D LLM '3D-PLOT-LLM' addresses object parts efficiently
Researchers have developed a new 3D multimodal large language model called 3D-PLOT-LLM that addresses the limitations of previous models in understanding and reasoning about object parts. Unlike prior approaches that re…
-
New datasets aim to advance AI-assisted CAD program generation
Researchers have introduced two new datasets aimed at advancing AI-assisted Computer-Aided Design (CAD). CADBench offers a unified benchmark with 18,000 samples across six families, five input modalities, and six metric…
-
New Monte Carlo method speeds 3D geometry processing and representation learning
Researchers have developed a novel Monte Carlo method to estimate the Dirichlet-to-Neumann (DtN) operator and its associated Steklov eigenmodes for geometry processing. This approach is significantly faster and more rob…
-
New benchmarks assess AI's CAD program generation capabilities
Two new benchmarks, CADBench and BenchCAD, have been released to evaluate AI's ability to generate Computer-Aided Design (CAD) programs from various inputs. These benchmarks aim to standardize the assessment of multimod…
-
MoCapAnything V2 enables end-to-end motion capture for arbitrary skeletons
Researchers have developed MoCapAnything V2, an end-to-end system for 3D motion capture from monocular video that accommodates arbitrary skeletons. This new framework integrates pose and rotation prediction into a singl…
-
New datasets aim to improve linguistic diversity and spatial alignment for embodied AI
Two new datasets aim to improve embodied AI research by addressing limitations in existing data. One paper, "Limited Linguistic Diversity in Embodied AI Datasets," audits current corpora and finds they often use repetit…