E-RayZer is a self-supervised 3D reconstruction system that takes unordered multi-view images as input and outputs estimated camera poses, target-view renders, Gaussian point clouds, and turntable videos. The system uses a transformer-based architecture to estimate camera poses and scene representations, then renders novel views using differentiable Gaussian splatting. It includes a Gradio demo interface for easy inference, along with training and evaluation scripts for the DL3DV dataset. The model is trained without ground truth camera poses or depth information, learning to reconstruct 3D scenes purely from multi-view image consistency.
This is a research codebase for the Geometric Context Transformer (GCT), a streaming monocular 3D reconstruction model accepted at ECCV 2026. The system processes RGB video frames in real-time (targeting ~20 FPS) to predict camera poses, depth maps, and 3D point clouds using a vision transformer with a novel paged KV-cache causal attention mechanism. It includes a full inference pipeline, interactive and offline renderers (based on Viser and Open3D/CUDA voxelization with octree LOD), and a comprehensive benchmark framework supporting multiple datasets (KTH, Oxford Spires, TUM, 7-Scenes, ETH3D, etc.) with trajectory, depth, and point cloud metrics. The codebase is predominantly application-oriented—providing a deployable model and evaluation harness—but incorporates significant data engineering components such as dataset preprocessing (e.g., Oxford Spires fisheye-to-pinhole conversion with CUDA visibility filtering), standardized BSS artifact storage, and reproducible conda-based environment setup. The core model architecture demonstrates advanced engineering practices: two-stream token layouts, sliding-window KV cache eviction with special-token preservation, 3D RoPE, and modular attention backends (FlashInfer, SDPA, custom causal). The benchmark framework is well-architected with pluggable methods and datasets, resumable execution, and static HTML report generation.
SPAG4D is a production-grade tool that converts 360° equirectangular panoramic photos into explorable 3D Gaussian Splat (3DGS) scenes. It integrates multiple depth estimation backends (DA360, DAP, PaGeR, SHARP, UniSHARP) to generate depth maps from a single panorama, then projects the points into standard 3DGS PLY format. The system provides a CLI, a FastAPI web server with a JavaScript frontend, and a Python API for programmatic use. Beyond the core conversion, it includes optional refinement pipelines to fill holes and improve scene quality: OmniRoam (WSL2-based video generation and distillation), ArtiFixer3D (Docker-based diffusion refinement), and a lighter geometric refinement pipeline using depth alignment and voxel aggregation. The architecture is modular with pluggable generators, comprehensive configuration dataclasses, and provenance tracking. While the tool is primarily an application, it leans heavily on research-grade depth models and refinement techniques, giving it a hybrid character. The codebase is well-structured with clear separation of concerns, reusable utilities (spherical projection, depth blending, scene filtering, geometric alignment), and extensive tests. It is actively maintained with recent commits, though it carries complexity from multiple backends and platform-specific dependencies (WSL2, Docker, CUDA).
WorldGen is a Python library and demo application for generating 3D scenes from text prompts or single images in seconds. It orchestrates a pipeline of state-of-the-art models: a fine-tuned FLUX diffusion model to generate 360° panoramic images, DA-2 for monocular depth estimation on equirectangular images, and optionally OneFormer for semantic segmentation and LaMa for inpainting to handle background occlusions. The 3D scene is produced as a Gaussian splatting point cloud (via a custom RGBD-to-Gaussian conversion) or as a textured mesh. An experimental mode uses Apple's Sharp model for higher-quality Gaussian splats by processing cubemap faces. The codebase demonstrates strong engineering practices for a research project: modular pipeline design, use of submodules for dependencies, custom diffusers pipelines with VAE tiling modifications for seamless panorama generation, and a Viser-based interactive viewer. While the core value is the integrated pipeline, several utility modules (equirectangular projection utilities, splat file handling) are cleanly isolated and reusable in other 360° or 3D vision projects.
SAIL-Recon is a feed-forward Transformer-based system for large-scale Structure-from-Motion (SfM) and visual localization. It processes input images or video to estimate camera poses, depth maps, and generate 3D point clouds without iterative optimization. The architecture combines a Vision Transformer backbone with an alternating attention aggregator for temporal modeling and specialized heads for camera, depth, and track predictions. The system includes both command-line and Gradio-based demo interfaces for end-to-end 3D reconstruction, along with evaluation scripts for standard benchmarks (TUM, 7Scenes, Tank&Temple, etc.). While research-oriented, it is packaged as a usable application with clear entry points and modular components.
Scal3R is a research codebase for CVPR 2026 Highlight paper on large-scale 3D reconstruction from unordered images using test-time training. The core contribution is a novel bidirectional attention mechanism with fast weight updates (TTT layers) that adapts during inference to improve reconstruction quality for large scenes. The repository provides an end-to-end inference pipeline: given a folder of images, it estimates camera poses, depth maps, and produces a colored point cloud. The architecture builds on a DINOv2 backbone and VGGT-based aggregator with custom attention heads, loop detection via SALAD, and pose graph optimization with submap management. The codebase is well-structured with separate modules for data loading, model definition, inference orchestration, and visualization tools. It includes fallback implementations for key components (e.g., Python-only loop solver, simple correlation-based loop detector) to improve portability, though GPU and CUDA are assumed. The repository is primarily a research artifact with clean utilities that could be extracted for other 3D vision projects. Evaluation code and dataset scripts are explicitly marked as not yet released, limiting reproducibility until those are provided.
GS-SDF is a research prototype from HKU-MARS that fuses LiDAR-augmented 3D Gaussian Splatting with a neural Signed Distance Field (SDF) for photorealistic rendering and high-fidelity surface reconstruction. The system ingests data from multiple sensor modalities (LiDAR, RGB-D, stereo) via a modular data parser hierarchy supporting Replica, FAST-LIVO2, KITTI, COLMAP, Oxford Spires, and other formats. It trains a neural SDF field with eikonal and curvature regularization, then initializes and refines 3D Gaussians guided by the SDF to produce view-dependent color and geometry. The pipeline includes real-time ROS visualization (RVIZ), offline mesh extraction via a custom CUDA marching cubes implementation, and evaluation scripts for image metrics (PSNR, SSIM, LPIPS) and structural metrics (Chamfer distance, F-score). The codebase is pre-release, with many TODOs and commented-out code, but exhibits strong modularity in its data loaders, loss functions, optimizer utilities, and coordinate transformations. While primarily a research codebase, it includes application-like features such as a ROS node with interactive keyboard commands and headless rendering modes.