VisionDepth3D v3.6 – Changelog

1) Upscaling & Interpolation – Massive Speed Boost

• Rewrote frames tab pipeline with threaded workers + queues.
• RIFE interpolation, ESRGAN upscaling, and FFmpeg writing now run concurrently instead of sequentially.
• Intelligent frame indexing + buffering preserves exact order while maximizing throughput.
• Render time reduced from 10 hours → ~1 hour on long clips.
• Result: creators can now upscale & interpolate full-length videos in a fraction of the time without crashes or dropped frames.

2) Depth Pipeline – Refinements & Blending

• Integrated Depth Blender prototype as a tab in main UI.
• Added early sliders/knobs for tuning blend weight between models.
• Improved handling of 16-bit depth outputs for richer disparity.
• Experimented with percentile clipping to suppress outliers without flattening depth.
• Result: cleaner separation between foreground and background, less “fuzz” in blended depth maps, and more consistent 3D parallax.
• Added Depth Anything V2 Giant model support
• Integrated FP16 half-precision toggle in main UI for faster inference and reduced VRAM usage (on supported GPUs).

3) HDR10 Support – Preservation & Metadata

• Fixed washed-out output issue when re-encoding HDR10 videos.
• Added flags to preserve:
  • 10-bit pixel format (yuv420p10le)
  • BT.2020 colorspace
  • PQ transfer curve (smpte2084)
  • Master-display / MaxCLL metadata
• Toggle option in UI: Preserve HDR10 Metadata.
• Result: HDR sources retain their original punch and dynamic range instead of flattening to SDR.

4) Experimental Live 3D (WIP)

• Added real-time 3D pipeline for external inputs (console, capture cards, webcams).
• Integrated Depth Anything v2 Small for on-the-fly depth generation (larger models can be swapped if GPU allows).
• Implemented VisionDepth3D method for stereoscopic conversion of live depth maps.
• Early end-to-end testing completed with capture → depth → stereo output loop.
• Performance tuning ongoing: fps optimization, GPU acceleration, and reduced latency paths.
• Expanded support for multiple input sources (video devices, capture cards, webcams).
• Result: playable console/video feeds in live 3D, showing strong potential for real-time applications.

5) General Fixes & Stability

Rendering

• Fixed: duplicate clip-window guard removed to prevent redundant early exits.
• New: Clip range UI — users can now set precise Start/End times for partial renders.
• New: Left-only and Right-only rendering modes now work directly (no post-split required).
• Optimized: hoisted per-eye resize + aspect ratio calculations out of the frame loop.
• Fixed: floating-window bar now scales correctly for single-eye renders.
• Optimized: DOF/Color grading checks moved outside the render loop to cut overhead.
• single-eye mode writes only the active eye, reducing memory work.
• Additional padding + edge reflection during stereo warp to reduce bleed-through.

UI & Error Handling

• Patched white-edge artifact issue caused by 16-bit depth normalization.
• Improved error handling in Preview GUI when models fail to load.
• Language files: synced new controls (HDR toggle, depth blender knobs, etc.) across all packs.
• All buttons and inputs use dark theme styling for a cleaner look.

Codec & Output

• Fixed: FFmpeg no longer forces -preset slow when a GPU codec is selected.
  • NVENC now uses encoder-appropriate flags (preset p5, rc vbr, cq).
  • CPU codecs continue to use CRF + preset for quality consistency.

6) UI & Workflow Enhancements

• Full 3D Generation Tab UI overhaul for a cleaner look

• Save/Load settings and presets with hotkeys

• Hotkey Import video & depth maps directly into the 3D workflow

• Reset in one click

• Jump straight to docs, bug reports, GitHub, or community links from the Help menu

• Result: smoother day-to-day workflow and better visual testing inside VD3D.

Summary:
v3.4 gave creators fine-grained depth & subject control. v3.5 brings cinematic polish with stabilized DOF, a true audio tool with sync + codec options, a GPU color grading suite, a stereo separation (IPD) adjustment for display comfort, and now ONNX pipeline + clip rendering support for faster experimentation. v3.6 focuses on depth blending refinements and proper HDR10 preservation, while also introducing precise clip-range rendering and direct Left/Right eye output for greater workflow flexibility.
This release tightens up edge handling, boosts stability, and continues to evolve the UI toward an all-in-one workflow.