Bi-directional image projection converter. Converts images between equirectangular and mercator projections.
Features a fast GPU and a slower CPU conversion implementations:
- The GPU version uses moderngl to transform the image using shaders and then saves it using Pillow.
- The CPU version simply uses Pillow to modify every pixel individually.
This project therefore contains working examples of Python and GLSL conversion code. Feel free to borrow them and translate them to your language/project.
- Install Python 3.11+
- Install requirements
pip install -r requirements.txt
python convert.py input output {m,e} [--nearest] [--cpu]Examples:
# Convert equi.png (equirectangular) to merc.png (mercator)
python convert.py equi.png merc.png m
# Convert merc.png (mercator) to equi.png (equirectangular)
python convert.py merc.png equi.png e| Argument | Description |
|---|---|
| input | The input file path. |
| output | The output file path. |
| {m,e} | Projection to convert the image to: "m" for mercator, "e" for equirectangular. |
| --nearest | Use nearest sampling for stretching that will occur due to change of aspect ratio. Only used by GPU rendering. |
| --cpu | Use the CPU rendering implementation. Much slower and doesn't support linear sampling. |
You can use the Converter class directly in Python. Check out the demo script to see how to run conversions from code.
The conversion process doesn't handle longitude angles of 85+ degress well. Converted images may display issues around the top and bottom borders.
If anyone wants to transform this into a functional, publishable package, feel free to fork the project and publish it. I don't have enough experience doing that and can't imagine many people will use this thing to make the extra effort worthwhile.