Changelog

For all practical purposes, this is the first proper release of depthflow.

• Move website to tremeschin.com domain, port contents into zensical
• Remove batch processing support for its many problems and better done in scripts
• Remove depth estimators cli (can use via library, image formats don't like u16, f32)
• Remove animation system, now in example scripts and documentation
• Remove Gradio WebUI for a future rework (looking for better libraries)
• Remove all upscalers wrappers (out of scope)
• Design an actual project logo rather than a placeholder.

• Limit imgui-bundle up to v1.6.3 per breaking changes

Changes ever since that would've been a v0.10, present in v1.0

• Fix internal resolution doubling bug before the final resize
• Fix FFmpeg command line interface options help missing
• Fix turbopipe.sync shouldn't be called when disabled
• Fix Docs videos missing, now hosted on GitHub user content
• Moved depth estimators to this package from shared library
• Swap logger library to DearLog from Loguru
• Make the inpainting mask color green
• Move to cyclopts cli library over typer

• Convert the project into snake case
• Overhauled the Readme and the WebUI layout and content
• Improvements to perceptual quality of the animation presets
• Add Upscayl as an upscaler option
• Add a command line interface for all upscalers and depth estimators
• Fixed drag and drop of files due new lazy loading logic
• Add stretch detection math on the shader for potential fill-in with gen ai
• Add colors filters (sepia, grayscale, saturation, contrast, brightness)
• Add transverse lens distortion filter (intensity, decay options)
• Overhaul animation system to be more flexible and reliable
• Reorganize website examples section into their own pages
• Cached depthmaps are now handled by diskcache for safer cross-process
• Refactor the shader for future include system external usage
• Simplify how the ray intersections are computed with ray marching
• Fix how projection rays are calculated, as steady, focus were slightly wrong
• Fix base scene duration is now 5 seconds

• Implement batch export logic within the main command line
• PyTorch is now managed in a top-level CLI entry point
• Many improvements to the website documentation: Quickstart, examples, and more
• Added Apple's DepthPro as an Depth Estimator option
• The exported video now properly follows the input image's resolution
• Loading inputs is now lazy, and only happens at module setup before rendering
• Improved the Readme with community work, quick links, showcase