VisionDepth/VisionDepth3D

What it solves

VisionDepth3D is an all-in-one 3D suite designed for creators to convert 2D images and videos into 3D stereo content for VR and cinema. It addresses the difficulty of creating high-quality depth-aware parallax shifting and stereo effects from flat media without requiring professional 3D production pipelines.

How it works

The software uses a hybrid approach combining AI-driven depth estimation and custom stereo logic. It leverages various AI models (such as Depth Anything v2, Marigold, and ZoeDepth) to generate depth maps, which are then processed through a GPU-accelerated stereo warping engine. This engine applies per-pixel parallax shifting, occlusion healing, and edge repair to create a 3D effect. Additionally, it includes tools for depth fusion (blending two depth sources), RIFE for FPS interpolation, and Real-ESRGAN for upscaling.

Who it’s for

It is aimed at individual creators, VR enthusiasts, and filmmakers who want to transform 2D content into 3D formats like Half-SBS, Full-SBS, VR180, and Anaglyph.

Highlights

• AI Depth Engine: Supports a wide range of models via PyTorch, TorchHub, Diffusers, and ONNXRuntime.
• Stereo Composer: Features subject-anchored convergence, depth shaping, and occlusion healing.
• Depth Blender: Allows the fusion of two depth sources for cleaner maps.
• Live 3D Sandbox: Provides real-time 2D-to-3D conversion for testing models and controls via camera or screen capture.
• Enhancement Tools: Integrated RIFE for high-FPS generation and Real-ESRGAN for 4K upscaling.
• Flexible Output: Supports multiple aspect ratios and containers (MP4, MKV, AVI) with hardware encoding (NVENC, AMF, QSV).