V3.8 (#81)

* UI Changes and new Encoder and Process options 

- Added buttons for encoder settings and processing options.
- Implemented multi-language support and tooltips for new dialog boxes.
- Adjusted preview image window size and video info layout to prevent window overflow.
- 3D tab columns now stack correctly when resizing the window on smaller screens.

* v3.8 3D Backend

- Reworked Auto Crop Black Bars to use first-frame detection with cached crop reuse.
- Prevents per-frame crop jitter and depth/frame misalignment.
- Improves stability for cinema content with subtle letterboxing.
- Keep Audio checkbox now respects the user-selected output container instead of forcing MP4.

* v3.8 - Depth Pipeline

### Depth Models

- Fixed ONNX model loading:
  - Distill-Any-Depth (inference resolution 518×518, batch size 8)
  - Video Depth Anything (inference resolution 512×288, batch size 8)
- Implemented LBM depth model (development version). Thanks to Aether for the implementation fix.
- Removed depth models from the dropdown that returned no `d_type`.
- Fixed Hugging Face model downloads and caching so zoo models consistently save inside the app `weights/` directory (no more extra `.cache` downloads).
- Updated Transformers image processor loading to prefer `use_fast=True` when available (with automatic fallback when unsupported).

### Depth Backend

- Implemented temporal smoothing in the depth pipeline to reduce flicker and improve temporal stability of depth map output.
- Packaged VisionDepth3D.exe with Distill-Any-Depth (ONNX), Video Depth Anything (ONNX), and Depth Anything v2 Giant weights.

* v3.8 External 3D

- Changed entry box to be easier to input 
- Console loading text changed

* Change log level for xFormers availability warning

* Change xFormers warning to debug log level

* v3.8 Frametool Pipeline 

- changed processor output text to match rest of system 
- Added SSResNet Model support for future implementation

* v3.8 Depth Blender 

- Standardized startup console messages to clearly reflect which subsystems are initializing (Torch, depth estimation, upscaler, external 3D pipeline, language, settings).

* v3.8 Preview GUI

- Changed size of frame to smaller size so Drop down doesnt get cropped to small

* v3.8

* v3.8

* v3.8

* v3.8

* v3.8

* v3.8

New Language

* Revise Changelog for VisionDepth3D v3.8

Updated changelog for version 3.8, detailing new features, improvements, and fixes across various tabs including Depth Estimation and 3D Render. Emphasized stability, model compatibility, and UI enhancements.

Author: VisionDepth

Changelog.md

@@ -1,204 +1,68 @@
-# VisionDepth3D v3.7 – Changelog
+# VisionDepth3D v3.8 – Changelog
 
 ---
 
-## 1) Live 3D Capture – Real-Time Overhaul
+## 1) Depth Estimation Tab
 
-### Audio Support Added
+### Depth Models
 
-* Added optional live audio passthrough monitor for external capture devices  
-* New flags:  
-  * `--audio-device "<device name>"`  
-  * `--audio-delay-ms <value>`  
-* Supports DirectShow and fallback to WASAPI  
-* FFplay-based audio pipeline integrated for low latency monitoring  
-* Audio sync delay control for VR headset streaming  
+- Fixed ONNX model loading:
+  - Distill-Any-Depth (inference resolution 518×518, batch size 8)
+  - Video Depth Anything (inference resolution 512×288, batch size 8)
+- Implemented LBM depth model (development version). Thanks to Aether for the implementation fix.
+- Removed depth models from the dropdown that returned no `d_type`.
+- Fixed Hugging Face model downloads and caching so zoo models consistently save inside the app `weights/` directory (no more extra `.cache` downloads).
+- Updated Transformers image processor loading to prefer `use_fast=True` when available (with automatic fallback when unsupported).
 
-### Color Channel Controls
+### Depth Backend
 
-* Added `--pixelshift-rgb` to properly route RGB output from CUDA warp kernel  
-* Added `--force-bgr-swap` fallback for cameras that provide BGR swapped frames  
-* Fixed purple and red tint issues on some capture devices  
-
-### Performance and FPS Optimization
-
-* Tuned live depth inference resolution defaults for smoother real-time processing  
-* Recommended real-time profile: `infer-w 448`, `infer-h 256`, depth FPS ~ 8 to 12  
-* Better GPU scheduling for live stereo warp when audio is active  
-* More stable stream with reduced stutter on high resolution capture input (1080p HDMI sources)  
-
-### Network Streaming Prep (Experimental)
-
-* Added headless running mode (`--no-preview`) to remove local display overhead  
-* Prepared architecture for browser-based SBS VR streaming  
-* In development: WebRTC path for synchronized audio + video to Quest browser  
-
-### General Stability
-
-* Improved diagnostic output and runtime logs  
-* Verified working flags for external capture cards (Elgato HD60 S+)  
-* Ensured fallback modes for low resource conditions  
-
----
-
-## 2) Initial Bugfixes & Issues Resolved (Live 3D Cleanup)
-
-### UI Settings Not Applying
-
-* GUI sliders and dropdowns weren’t being respected when starting live capture  
-* Fixed runtime sync so all selected GUI settings apply properly  
-
-### Manual CLI vs GUI Discrepancy
-
-* Runtime ignored resolution, backend, FPS, and other settings when launched via GUI  
-* Now fixed, all selected parameters propagate as expected  
-
-### Video Tint Correction
-
-* Purple/red tint caused by incorrect channel routing  
-* Fixed with `--pixelshift-rgb` and `--force-bgr-swap` fallbacks  
-
-### Capture Failures (Webcam/HDMI)
-
-* Fixed `no frames arriving` error by enforcing `--fourcc MJPG` and proper MSMF backend handling  
-
-### No Audio Output
-
-* Live mode initially had no sound output  
-* Added full audio routing support for monitoring devices (DirectShow/WASAPI)  
-
-### FPS Performance Bottlenecks
-
-* Live stereo + depth ran around 6.5–7 FPS at first  
-* Now tuned with smaller inference size, better scheduling, and CUDA-side warp boost  
-
----
-
-## 3) Floating-Window and Stability Fixes
-
-### Dynamic Floating Window (DFW) Logic
-
-* Fixed undefined `zero_parallax_offset` references and long term stabilization drift.
-* Rebuilt the DFW into an asymmetric, side only window that masks a single edge (left or right) based on the dominant parallax direction.
-* Added a minimum parallax threshold so the window stays completely off when depth is very close to the screen plane.
-* Computes bar width as a blend of parallax magnitude and how far the tracked subject is from mid depth, then clamps it to a small fraction of the per eye width for a subtle mask.
-* Uses `FloatingWindowTracker` to smooth the horizontal offset over time and reduce small frame to frame jitters.
-* Uses `FloatingBarEaser` to ease the bar width in and out so it expands and collapses gradually instead of popping on and off.
-* Supports both soft faded masks and solid black cinema bars through a single toggle, with faded mode as the default for VR and monitor playback.
-* Result: a more invisible and cinema friendly floating window that reduces window violations, keeps edges clean, and stays out of the viewer’s way.
-
-### Frame Jitter and Temporal Stability
-
-* Fixed flickering, depth “breathing,” and jitter caused by unsmoothed convergence and subject tracking.
-* Applied `SubjectDepthEMA`, `DepthPercentileEMA`, and `ConvergenceEMA` for improved temporal stability.
-* Added a global `ShiftSmoother` to stabilize foreground, midground, and background parallax.
-* Result: consistent parallax motion across frames, no more in-and-out depth pulsing, and a much cleaner, more comfortable stereo flow.
-
-### Auto-Crop Black Bar Logic
-
-* Fixed crop mis detections during fade ins and fade outs.
-* Added a mean brightness guard so black bar detection does not update on very dark transition frames.
-* Added per frame re evaluation when black bar height changes significantly.
-* Result: letterboxed content (2.35:1 and similar) now auto crops reliably without vertical drift.
+- Implemented temporal smoothing in the depth pipeline to reduce flicker and improve temporal stability of depth map output.
+- Packaged VisionDepth3D.exe with Distill-Any-Depth (ONNX), Video Depth Anything (ONNX), and Depth Anything v2 Giant weights.
 
 ---
-## 4) Unified Depth Pipeline Upgrade & Platform Stability Improvements
 
-v3.7 introduces the largest Depth tab upgrade yet, combining:
+## 2) 3D Render Tab
 
-- **GPU backend support across NVIDIA, AMD, and Apple Silicon**
-- **FFmpeg codec selection for hardware-accelerated depth exports**
-- **Pause / Resume / Cancel controls** for long video renders
+### UI Fixes
 
----
-
-### GPU Backend & Platform Support (CUDA / ROCm / MPS / CPU)
-
-* Full rewrite of device detection — CUDA is no longer assumed  
-* Automatic selection of best available compute backend  
+- Added buttons for encoder settings and processing options.
+- Implemented multi-language support and tooltips for new dialog boxes.
+- Adjusted preview image window size and video info layout to prevent window overflow.
+- 3D tab columns now stack correctly when resizing the window on smaller screens.
 
-#### Supported Depth Inference Backends
-- **CUDA** — NVIDIA GPUs  
-- **ROCm** — AMD GPUs  
-- **MPS** — Apple Silicon GPUs  
-- **CPU fallback** when no GPU is detected  
+### 3D Backend
 
-#### Benefits
-* Prevents CPU-only fallback on capable GPUs  
-* Removes CUDA-only links that caused AMD/macOS crashes  
-* Core foundation for **Linux** deployment moving forward
-
-> Result: VD3D depth inference now runs properly on **NVIDIA, AMD, Apple, and CPU-only** systems.
+- Reworked Auto Crop Black Bars to use first-frame detection with cached crop reuse.
+- Prevents per-frame crop jitter and depth/frame misalignment.
+- Improves stability for cinema content with subtle letterboxing.
+- Keep Audio checkbox now respects the user-selected output container instead of forcing MP4.
 
 ---
 
-### FFmpeg Codec Selection for Depth Video Rendering
-
-* New Video Codec dropdown added to the Depth Tab  
-* Hardware and software encoding now user-selectable  
-
-#### Supported Encoders
+## Frametool Backend
 
-**NVIDIA NVENC**
-- `h264_nvenc`, `hevc_nvenc`, `av1_nvenc`
-
-**AMD AMF**
-- `h264_amf`, `hevc_amf`, `av1_amf`
-
-**Intel QSV**
-- `h264_qsv`, `hevc_qsv`, `vp9_qsv`, `av1_qsv`
-
-**CPU Fallback**
-- `libx264`, `libx265`, `libaom-av1`, `libsvtav1`, `mp4v`, `XVID`, `DIVX`
-
----
-
-#### Safety & Compatibility Enhancements
-* Fixes XVID encoding failures on AMD/Intel systems  
-* Unifies codec support with 3D Converter & FrameTools  
-* AV1 warning system alerts users when OpenCV cannot decode  
-* Built to allow full FFmpeg pipeline in next release  
+- Reworked Frametool backend to support SSResNet models for feature model integration.
 
 ---
 
-### Depth Pipeline Control System
+## Console Improvements
 
-* **Pause**, **Resume**, and **Cancel** depth rendering  
-* Real-time resource management during pauses  
-* Safe termination prevents file corruption  
-* Clear progress status states:
-  - Running  
-  - Paused  
-  - Canceling  
-  - Completed  
-
-> Enables fast workflow adjustments and prevents wasted GPU/CPU time.
+- Standardized startup console messages to clearly reflect which subsystems are initializing (Torch, depth estimation, upscaler, external 3D pipeline, language, settings).
+- Unified compute device reporting across pipelines for consistent and clearer console output.
+- Suppressed optional xFormers dependency warning on startup.
+- Prevented duplicate language loading during settings restore.
 
 ---
 
-## 5) 3D Pipeline & UX Polish
-
-- Added a **Keep Original Audio** checkbox to optionally copy the source video’s audio into the final 3D render (no re-encoding).
-- Hooked a new **image-based 3D pipeline** directly into the main 3D renderer for single-frame conversions.
-- Wired the **Mode** selector so it cleanly switches between **Single**, **Batch**, and **Image** workflows.
-- Implemented an automatic **3D filename suffix system** so exports are labeled by format and eye mode  
-  (for example: `_LRF_Full_SBS`, `_LRF_Half_SBS`, `_VR`, `_Anaglyph`, `_Interlaced`, `_LRF_Left`, `_LRF_Right`).
-- Reviewed and cleaned up **multi-language labels and tooltips** across all supported locales.
-
-## 6) Depth Blender – Live Preview & Scrubber
-
-- The **Depth Blender** tab now has a fully working live preview, showing the V2 base map and the blended result side by side.
-- All blend parameters (white strength, feather blur, CLAHE, bilateral filters) now update the preview in real time, making it easier to dial in a clean, stable depth mix before running a full batch.
-
-
 ## Summary
 
-v3.7 focuses on stabilizing **Live 3D Capture**, fixing sync and GUI issues, and refining floating window behavior.  
-Combined with unified **GPU backend support** and **hardware-accelerated codecs** for depth rendering, VD3D is now significantly more stable across:
+v3.8 focuses on stabilizing depth estimation, improving model compatibility, 
+and refining the 3D Render tab UI with better layout behavior, clearer diagnostics, and improved localization support.
 
-- NVIDIA systems  
-- AMD ROCm systems  
-- Apple Silicon (M1/M2/M3)  
-- CPU-only environments
+> Back up your `weights/` and `presets/` folders before uninstalling v3.7.
+> Then run VisionDepth3D_Setup_Downloader to download the official
+> VisionDepth3D v3.8 Windows installer and required `.bin` files.
 
-These improvements build the foundation for **Linux** and broader cross-platform support.  
+> (Optional but recommended) Clear the Hugging Face cache to free space and
+> avoid duplicate model downloads:
+> `C:\Users\YOUR_USERNAME\.cache\huggingface`