So the version 1.1.0 is live. Even though there was a lot of code changes in the background, normally the experience should not be too different compared to version 1.0.4.
You will be asked if you want to delete the models installations when first launching the program. If you want to go back to version 1.0.4 at some point, you should say no to this prompt but be aware that models installations are not compatible between both versions.
Here is a list of the main changes:
All Modes:
- Models settings are now saved.
- 180° and 360° screen shapes are now selectable. To be used with 180° or 360° videos, some can be found there (they need to be downloaded): https://www.youtube.com/channel/UCzuqhhs6NWbgTzMuM09WKDQ
- UI buttons are slightly smaller to allow for more options to be displayed.
Screen Capture Mode:
- Vertical cropping as been added to reduce the vertical size of the area sent to the depth estimation
Video Mode:
- Introduction of the video mode, which allow you to select a video file to be read.
- .mkv format is not supported as the Unity video player doesn't handle that format.
- Moving the mouse while playing a video will show the Deep Desktop interface and allow you to make changes to settings.
- Caching options allow saving depth estimations for later watching of videos.
- NOTE: Performance doesn't scale well with high resolutions, so 4K videos might be better being read using screen capture.
Extensions:
- So the purpose of extensions is to allow programs that do other things than Depth estimation to be run in the background and provide other features. For now, I did only put PaddloOCR for text extraction and both Deep Translator and PyKakasi (for japanese kanji to kana "translation") for translation extensions. To use the translation extensions you need to have a text extraction plugin installed. Settings for translations need to be adjusted in the Extensions tab in the Screen Capture or Video menu.
- Extensions can be turned on in the Extensions menu, they are off by default after having been installed
Models:
- Added the Depth-AnythingV2-Metric model
- Fixes to the Marigold model as it has many issues when rendering the result