Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/universaldatatool/autoseg
WebAssembly Realtime Image Segmentation
https://github.com/universaldatatool/autoseg
image image-processing image-segmentation webassembly
Last synced: about 2 months ago
JSON representation
WebAssembly Realtime Image Segmentation
- Host: GitHub
- URL: https://github.com/universaldatatool/autoseg
- Owner: UniversalDataTool
- License: mit
- Created: 2020-06-16T22:35:30.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2023-01-06T12:32:15.000Z (about 2 years ago)
- Last Synced: 2024-10-29T18:45:47.009Z (3 months ago)
- Topics: image, image-processing, image-segmentation, webassembly
- Language: C++
- Homepage: https://universaldatatool.com
- Size: 2.58 MB
- Stars: 6
- Watchers: 3
- Forks: 2
- Open Issues: 23
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# autoseg | Automatic Fast WebAssembly Image Segmentation
> Are you looking to convert UDT annotations into png masks? Use [udt-to-png](https://github.com/UniversalDataTool/udt-to-png)
autoseg segments images from a list of points and polygons containing
classifications. Autoseg was originally created by [Severin Ibarluzea](https://twitter.com/seveibar) for usage with the [Universal Data Tool](https://github.com/UniversalDataTool/universal-data-tool).
## Features
- Simple API
- WebWorker background computation
- Deterministic, suitable for mask compression
- Fast. WebAssembly compiled from state of the art C++ implementations of SLIC superpixeling and graph cut
- Fully compliant with the [Universal Data Tool](https://github.com/UniversalDataTool/universal-data-tool) and [UDT Format](https://github.com/UniversalDataTool/udt-format)
## Example Inputs -> Output
### Input Image
### Input Polygon
### Output Mask
## CLI
Make sure to format your file in the [\*.udt.json format](https://github.com/UniversalDataTool/udt-format).
```bash
npm install -g autoseg
# This will output a bunch of image pngs for the masks of a udt file
autoseg some_file.udt.json -o output-directory
autoseg --help
# Usage: autoseg path/to/dataset.udt.json -o output-masks-dir
#
# Options:
# --help Show help [boolean]
# --version Show version number [boolean]
# --use-sample-number Use the number of the sample as the mask filename
# (mask0001.png, etc.)
# --output-dir, -o Output directory for masks [required]
```
## API
```javascript
const autoseg = require("autoseg") // OR require("autoseg/node")
await autoseg.loadImage({ data: imData, width: 320, height: 249 })
// Points/polygons are in the UDT Shape format
// https://github.com/UniversalDataTool/udt-format
const maskImageData = await autoseg.getMask([
{ regionType: "point", x: 50, y: 50, cls: 0 },
{ regionType: "point", x: 150, y: 150, cls: 1 },
])
// If you have a canvas, you can draw the image of the mask
canvasContext.putImageData(maskImageData, 0, 0)
// NOTE: nodejs doesn't have builtin support for ImageData, but you
// can use the returned data in a similar way, it's an object with
// { data: Uint8ClampedArray, width: number, height: number }
```
## Configuring
```javascript
await autoseg.setConfig({
maxClusters: 1000,
classNames: ["dog", "cat"],
classColors: [0xffffffff, 0xff000000],
})
```