depth-anything-v2

Official code for the paper: Depth Anything At Any Condition

an easy way to create Abrasion/Scratch/Hightlight Holography from 2D images/pictures;轻易从平面图像/照片创建刮擦全息路径

Optimizing Monocular Depth Estimation with TensorRT: Model Conversion, Inference Acceleration, and 3D Reconstruction

Tensorrt codebase to inference in c++ for all major neural arch using onnx

M2Retinexformer: Multi-Modal Retinexformer for Low-Light Image Enhancement [IEEE ICIP 2026] is a novel framework that extends Retinexformer by incorporating depth cues, luminance priors, and semantic features within a progressive refinement pipeline.

Ready-to-deploy models including Segment Anything 3, Depth Anything 2 and Gemma.

Turn any video into depth maps, pose skeletons and 478-point face clouds for AI motion control (Seedance 2.0 / Kling / Runway) — 100% in-browser, no install, no upload.

[AI POC - STATUS: WORKING AT FULL SPEED, BUT POOR HALOING AND EDGE ALIGNMENT] DirectShow filter to convert 2D videos to 3D SBS for common media players (which support external filters) in real-time. Made with Claude.

Transform 2D images into 3D worlds using deep learning. This small project explores monocular depth estimation and 3D point cloud generation from a single image.

A pipeline to identify optimal picking surfaces on packages and estimate their surface normals for robotic manipulation. by using Depth Anything V2 and Segment Anything Model - SAM.

Comparative analysis of monocular depth estimation methods (ResNet-50, frozen Stable Diffusion UNet, I-JEPA, SD+I-JEPA fusion, DepthAnything V2) with robustness evaluation under fog, blur, and low-light on NYU Depth V2.

Convert images, PNG sequences and videos into MagicEye autostereograms. AI depth via MiDaS / Depth Anything V2. PySide6 UI inspired by X.

Fine-tuned version of Multi-Model-Monocular-Depth-Estimation-for-ROS-2 — ROS 2 (Humble) package for YOLO26 Depth and Depth Anything V2, using checkpoints fine-tuned on NYU Depth V2 and ScanNet (hosted on Hugging Face) instead of base pretrained weights.

Comparison of Depth Anything V2 and MiDaS for indoor monocular depth estimation and obstacle detection on NYU Depth V2.

MassingPro is a lightweight, zero-compute architectural utility that converts 2D facade photographs into UV-mapped 3D context models. It bypasses complex 3D topology by generating 16-bit displacement and normal maps, allowing for high-fidelity architectural visualization at zero CAD modeling cost.

Monocular-only autonomous indoor navigation for low-cost search-and-rescue drones: appearance-free door entry and TRISTAR tri-signal stair climbing on a DJI Tello.

Real-time monocular depth estimation and GPU visual effects using Depth Anything V2, PyTorch, CUDA, OpenGL, and GLSL. Converts webcam video into depth-aware fog, blur, lighting, and heatmap effects, with a Streamlit dashboard for live visualization, controls, telemetry, and performance monitoring.

ROS 2 (Humble) package wrapping three interchangeable monocular depth estimation models — YOLO26 Depth, UniDepth V2, and Depth Anything V2 — using stock pretrained weights, with a fine-tuned companion repo available for NYU Depth V2/ScanNet-adapted checkpoints.

Shared offline evaluation toolkit for depth estimation checkpoints — YOLO26-Depth and Depth Anything V2, PyTorch and OpenVINO, ScanNet benchmarks and local lab-generalization tests, one consistent metric pipeline throughout.