2D-3D CONVERSION USING DEPTHANYTHINGV2 AND VIDEO-DEPTHANYTHING

Download AI DepthStickV51 (7.8 GByte): DepthStick_V51

How to use it:

• V51 Update 17.10.2025: Permanent pause checkbox
• V51 Update 17.10.2025: The image directory may also contain non-jpg files.
• Supports up to nVidia RTX 50xx graphic cards
• Using nVidia graphic cards allows significantly faster processing.
• File and folder names must not contain umlauts or other special characters.
• Only unzip to a directory with write permissions, e.g. not to C:\Program Files!
• For image conversion: Only the images to be converted may be located in the input directory (no other files or dirs).
• For video conversion: The input directory may only contain one mp4 video file to be converted.
• Maximum lenght for video conversion: around 1800 frames full-HD (with 32 GByte RAM, maybe more with more RAM).
• Recommendation if unzipping (allow a few minutes!) does not work right away: 7-zip.org
• AI project pages: DepthAnythingV2 and Video-DepthAnything



• CUDA is 10 times faster for images and 1000 times faster for videos!
• When CUDA is enabled on systems without CUDA, there is a fallback to CPU.
• When CUDA is activated on systems with insufficient VRAM, there is a fallback to the smaller model or to the CPU.
• Minimum CUDA VRAM for video conversion with small model: 6 GB (starting from nVidia RTX 1060)
• Minimum CUDA VRAM for video conversion with large model: 16 GB (starting from nVidia RTX 4060Ti)


• For editing image depth masks, I recommend the Depth Map Editor by Bernd Paksa.

Performance comparison for an i7-14700K/RTX 4060Ti (16GB VRAM) computer:

Images/Video           Model   Device           Time
====================================================
Images MonaLisa        Small   CPU            1400ms
Images MonaLisa        Small   CUDA            140ms
Images MonaLisa        Large   CPU            5000ms
Images MonaLisa        Large   CUDA            400ms

Video  Rollercoaster   Small   CPU           2:04:52
Video  Rollercoaster   Small   CUDA               8s
Video  Rollercoaster   Large   CPU    (far too long)
Video  Rollercoaster   Large   CUDA             4:09

Frames (Mühlenbilder)  Small   CPU             31:45
Frames (Mühlenbilder)  Small   Cuda               2s
Frames (Mühlenbilder)  Large   CPU    (far too long)
Frames (Mühlenbilder)  Large   Cuda             1:04

To the tutorials on Image processing
To the tutorials on Video processing
To the tutorials on Rating & Sorting

© Gerhard P. Herbig, 2025,     back to cosima homepage