Release v2.3.9

Key Updates

• Add torch.compile and CuDNN support for AMD ROCm7 supported GPUs with 1.5~2x performance

• Add CoreML for Apple Silicon (MPS) devices with 2x performance

• Enhance the TensorRT performance and compatibility for NVIDIA GPUs (Support more models including DA3MONO, DA3METRIC)

• Add XPU Pytorch backend with OpenVINO optimizer for Intel GPUs. (Better than DirectML, requires 11th Gen and newer iGPUs and dGPUs)

• Enable automatic inference optimizer options for supported hardware
• Enable GPU OpenCL acceleration for the Screen Capture process with lower CPU usage
• Change RTMP server backend with SRT protocol on Windows with lower latency on supported RTSP and WebRTC clients
• Correct depth calculation and normalization for DA3 and metric models
• Update ROCm7 hardware installation script with official ROCm7.2 (9700/9070/9060/7900/7800/7700/840M/860M/880M/890M/8060s) and TheRock 7.1.2 for (Almost all RDNA2/3/4 GPU/APUs)

Performance Benchmark

Desktop2Stereo Performance Win/Mac

Alternative Download Link

• Quark NetDrive
  Access code: 1vcn
• Baidu Netdisk
  Access code: mr64

Release v2.3.8

Key Updates

• Optimize running process of Stereo Viewer, add proper support for [Lossless Scaling]. (https://store.steampowered.com/app/993090/Lossless_Scaling/) under Local Viewer and RTMP Streamer modes.
• Add Depth Map display mode to view real time depth video.
• Hide the viewer window on Windows when running RTMP Streamer mode (You can still show the window by enabling the Lossless Scaling Support).
• Update DA3 with v1.1 models: DA3NESTED-GIANT-LARGE-1.1, DA3-LARGE-1.1, DA3-GIANT-1.1.
• Update ROCm7.2 support for RX 6000 GPUs (RDNA2) on Windows based on ROCm7 library builds from guinmoon/rocm7_builds.
• Auto disable TensorRT for incompatible depth models (NVIDIA GPU).
• Enable automatic deactivation of Recompile TensorRT once user check and recompile (NVIDIA GPU).
• Enable automatic disable FP16 for DirectML devices
• Optimize the performance with better GPU usage efficiency especially for Legacy Streamer mode
• Fix wrong depth results for Depth-Pro

Alternative Download Link

• Quark NetDrive
  Access code: 1vcn
• Baidu Netdisk
  Access code: mr64

Release v2.3.7

Key Updates

• Update GUI with an option to set default target (virtual) monitor for SBS output.
• Add a direct run_main script to run Desktop2Stereo without GUI like legacy v1 version.
• Add ROCm7 support on Windows for AMD 7000/9000/Ryzen AI (Max)/etc. Series GPU , check compatibility here: https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html
• Add Support Depth-Anything-3 models (Beta: may have issues of sky segmentation or HUD artifacts)
• Slightly improve performance and better hole-filling for Local Viewer

Alternative Download Link

• Quark NetDrive
  Access code: 1vcn
• Baidu Netdisk
  Access code: mr64

Release v2.3.6

Key Updates

• Besides Windows, add the RTMP Streamer for both Mac and Linux with Audio streaming supports.
• Display down streaming Protocol URLs (RTMP/RTSP/HLS/HLS M3U8/WebRTC) directly in GUI (supported by MeidaMTX) .
• Update the activation of WindowsCapture process by Win+Tab on Windows.
• Move RTMP's viewer to next screen by default.
• Enable Vsync which may fix glitch.
• Optimize the 3D resolution and strength.
• Correct the fullscreen mode on MacOS.
• Remove some incompatible depth estimation models.
• Enhance the video smoothness and stability of MJPEG streamer.
• Enhance the stability of RTMP streamer.
• Replace PyAudio library with SoundDevice for better OS compatibility.
• Avoid some compiling failures when enable torch.compile and TensorRT on NVIDIA GPUs.
• Fix the Runtime Error of Legacy MJPEG Streamer mode on Intel Graphics GPUs.

Tips

• Virtual Display software such as
  1. Virtual Display Driver for windows
  2. BetterDisplay
     is highly recommended is highly recommended if you only have one monitor, better to set the same resolution as your main monitor (W, H), if you prefer Full-SBS mode, please set the width double for the virtual display (2W, H).
• All Stream Protocol URL under RTMP Streamer mode are valid while running, so no need to restart the RTMP Streamer. Recommend to use HLS for most VR devices, and HLS H3U8 for VLC App with AR devices.
• Quick press space/enter twice to toggle fullscreen mode on MacOS.
• For Mac, a Depth Resolution around 196 is recommended.

Alternative Download Link

Quark NetDrive
Access code: 1vcn

Release v2.3.5

Key Updates

• Optimize depth prediction performance, ~5FPS increase for 4K with DXCam based screen capture wincam.
• Add the Windows Capture Python as an screen capture option with low CPU temperature
• Add a RTMP Streamer for Windows (powered by FFMPEG and MeidaMTX) with Audio streaming support, latency ~3s
• Add a local 3D monitor mode for directly show SBS/TAB view on one monitor (not supported streaming)
• Add 2 new options to lock 16:9 ratio and keep local viewer aspect ratio
• Add a new MJPEG streamer
• Minor Bug fix for Video Depth Anything and Depth Resolution
• Add install scripts end with "0" for those who can directly access official pypi and pytorch sites

Tips

1. DirectML devices may have better and stable performance with FP16 off
2. New RTMP Streamer needs to keep the local viewer run at the backend
3. The Legacy Streamer sometimes may have better performance than the new one
4. Please make sure to enable the Windows StereoMix device in your system Audio Device settings

Release v2.3.4

Key Updates

• Add torch.compile and TensorRT inference optimizers for NVIDIA GPU.
• Enhance streamer performance for DirectML devices like AMD GPU.
• Add Video-Depth-Anything models to model list
• Enhance streamer stability and performance under window mode
• Update the endpoint as dropdown list in GUI
• Upgrade to Python3.11 and PyTorch 2.7.1 for better performance
• Add auto-update script

Release v2.3.3

Key Updates

• Add FPS display to local viewer, can toggle by "F" key
• Add Depth strength display to local viewer when adjusting using ⬆️ and ⬇️ arrow keys.
• Enhance the depth stability of 3D output, reduce flickering
• Use new depth postprocessing method
• Add foreground scale option for better depth effects, remove edge dilution (induce artifact)

Release v2.3.2

Key Updates

• Enable dedicated depth resolution lists for each model
• Enhance the streamer FPS for dml devices
• Add optional FPS display on the streaming video
• Further enhance depth strength
• Make MJPEG streamer smoother

Release v2.3.1

Key Updates

• Add the App Window capture mode
• Add a LAN streamer running mode
• Reduce edge artifacts and improve 3D effects, add Anti-aliasing and Edge dilation
• Optimize performance for 4K (~20FPS) and 1080P (>40FPS) with GPU on Windows
• Add adaptive MacOS cursor rendering under different HiDPI
• Add more hugging face Depth models
• Strengthen 3D effect

Release v2.2

Key Updates

• Add cursor capture from PyObjC API for Mac
• Optimize installation requirements
• Allow manual customize output resolution
• Adjust App icon size for Mac/Windows
• Add Universal Portable version for Windows
• Remember the GPU device when reset
• Correct FPS calculation
• Fix DTYPE mismatch issues for Depth-Pro and Intel-dpt-Large
• Fix low FPS bug due to depth resolution input