bgremover-onnx 2.0.0__tar.gz

bgremover_onnx-2.0.0/.gitignore

@@ -0,0 +1,50 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+dist/
+*.egg-info/
+.eggs/
+pip-wheel-metadata/
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# Testing and tooling caches
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+.coverage.*
+htmlcov/
+coverage.xml
+
+# Model weights are downloaded into a cache, never committed
+*.onnx
+models/
+
+# Sample and scratch outputs
+*_nobg.png
+outputs/
+uploads/
+temp/
+
+# Environment
+.env
+.env.*
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Editors
+.vscode/
+.idea/
+*.swp
+*.swo
+*~

bgremover_onnx-2.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Asim Mahmudov
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

bgremover_onnx-2.0.0/PKG-INFO

@@ -0,0 +1,290 @@
+Metadata-Version: 2.5
+Name: bgremover-onnx
+Version: 2.0.0
+Summary: Remove image backgrounds with ONNX segmentation models. Library, CLI and HTTP API.
+Project-URL: Homepage, https://github.com/asimmakhmudov/bgremover
+Project-URL: Issues, https://github.com/asimmakhmudov/bgremover/issues
+Project-URL: Documentation, https://github.com/asimmakhmudov/bgremover#readme
+Project-URL: Source, https://github.com/asimmakhmudov/bgremover
+Author: Asim Mahmudov
+License: MIT
+License-File: LICENSE
+Keywords: background-removal,birefnet,isnet,onnx,segmentation,u2net
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Multimedia :: Graphics
+Classifier: Topic :: Scientific/Engineering :: Image Processing
+Requires-Python: >=3.10
+Requires-Dist: numpy>=1.24
+Requires-Dist: onnxruntime>=1.16
+Requires-Dist: pillow>=10.0
+Provides-Extra: api
+Requires-Dist: fastapi>=0.110; extra == 'api'
+Requires-Dist: python-multipart>=0.0.9; extra == 'api'
+Requires-Dist: uvicorn>=0.27; extra == 'api'
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == 'dev'
+Requires-Dist: fastapi>=0.110; extra == 'dev'
+Requires-Dist: httpx>=0.27; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: python-multipart>=0.0.9; extra == 'dev'
+Requires-Dist: ruff>=0.5; extra == 'dev'
+Requires-Dist: twine>=5.0; extra == 'dev'
+Requires-Dist: uvicorn>=0.27; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# bgremover
+
+[![PyPI](https://img.shields.io/pypi/v/bgremover-onnx.svg)](https://pypi.org/project/bgremover-onnx/)
+[![Python](https://img.shields.io/pypi/pyversions/bgremover-onnx.svg)](https://pypi.org/project/bgremover-onnx/)
+[![CI](https://github.com/asimmakhmudov/bgremover/actions/workflows/ci.yml/badge.svg)](https://github.com/asimmakhmudov/bgremover/actions/workflows/ci.yml)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://github.com/asimmakhmudov/bgremover/blob/master/LICENSE)
+
+Remove image backgrounds with pretrained ONNX segmentation models — as a Python
+library, a command line tool, or an HTTP API.
+
+It runs U²-Net, IS-Net and BiRefNet with its own inference code: numpy and
+Pillow for the image math, onnxruntime for the model. There is no `rembg`
+dependency. (The weight files are hosted on rembg's releases page, which is the
+only thing taken from that project.)
+
+```python
+from bgremover import remove
+
+with open("photo.jpg", "rb") as fh:
+    png = remove(fh.read())
+```
+
+## Install
+
+With pip:
+
+```bash
+pip install bgremover-onnx            # library + CLI
+pip install "bgremover-onnx[api]"     # also the HTTP API
+```
+
+With [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add bgremover-onnx                 # add to a uv project
+uv add "bgremover-onnx[api]"
+uv pip install bgremover-onnx         # into the active environment
+uvx --from bgremover-onnx bgremover remove photo.jpg   # run the CLI without installing it
+uvx --from "bgremover-onnx[api]" bgremover serve   # run the API without installing it
+```
+
+The package is published on PyPI as `bgremover-onnx`; you still import it as
+`bgremover` and run the `bgremover` command.
+
+Python 3.10 or newer. Model weights are downloaded on first use and cached in
+`~/.cache/bgremover` (override with `BGREMOVER_HOME`).
+
+## Project structure
+
+```
+bgremover/
+├── src/bgremover/
+│   ├── __init__.py        Public API, resolved lazily (remove, Remover, MODELS, ...)
+│   ├── models.py          Model registry: URL, SHA-256, input size, normalisation
+│   ├── download.py        Download, hash-check and cache weights (stdlib only)
+│   ├── session.py         Remover: owns the ONNX session, runs the pipeline
+│   ├── processing.py      Pure image math: preprocess, postprocess, refine, cutout, trim
+│   ├── images.py          Decode any input (bytes / path / PIL / numpy), encode output
+│   ├── cli.py             `bgremover` command: remove, models, download, serve
+│   └── server/            Optional HTTP API (installed with the [api] extra)
+│       ├── app.py         FastAPI app, routes, error handling, bounded worker pool
+│       ├── config.py      Settings read from BGREMOVER_* environment variables
+│       └── fetch.py       SSRF-safe image fetching for the URL route
+├── tests/                 pytest suite (fake ONNX session; real models with -m model)
+├── .github/workflows/
+│   ├── ci.yml             Lint, test on 3.10-3.13, build the package, smoke-test Docker
+│   └── publish.yml        Publish to PyPI when a GitHub release is published
+├── Dockerfile             Two stages: bake weights in, then a slim non-root runtime
+└── pyproject.toml         Package metadata, extras, ruff and pytest config
+```
+
+How a request flows:
+
+```
+input (bytes / path / PIL / numpy)
+  -> images.load_image          decode to an RGB PIL image
+  -> processing.preprocess      resize to the model's input, normalise, NCHW tensor
+  -> onnxruntime                run the segmentation model (session.Remover)
+  -> processing.postprocess     scale the mask back to the image size
+  -> processing.refine_mask     optional: harden + feather the edge
+  -> processing.cutout / trim / apply_background
+  -> images.encode_image        same type out as came in
+```
+
+The core library never imports the `server` package, and `processing.py` does no
+I/O, so each layer can be tested on its own.
+
+## Library
+
+`remove()` returns the same type you give it: bytes in, bytes out; a PIL image
+in, a PIL image out; a numpy array in, a numpy array out; a path in, a PIL image
+out.
+
+```python
+from bgremover import remove, Remover
+
+png_bytes = remove(jpeg_bytes)                     # -> bytes (PNG)
+image     = remove("photo.jpg")                    # -> PIL.Image (RGBA)
+array     = remove(numpy_rgb)                      # -> numpy array (RGBA)
+
+# Reuse one model across many images - the session is built once.
+remover = Remover("isnet-general-use")
+for path in paths:
+    remover.remove(path).save(path.with_suffix(".png"))
+```
+
+Options for `remove()` / `Remover.remove()`:
+
+| Option | Default | Meaning |
+| --- | --- | --- |
+| `only_mask` | `False` | Return the grayscale mask instead of the cutout |
+| `refine` | `False` | Harden the confident regions and feather the edge |
+| `foreground_threshold` | `240` | With `refine`: at or above this is fully opaque |
+| `background_threshold` | `15` | With `refine`: at or below this is fully transparent |
+| `erode_size` | `0` | With `refine`: pull the edge in by N pixels |
+| `blur_radius` | `1.0` | With `refine`: feather radius |
+| `background` | `None` | Composite onto a colour (`"white"`, `"#ff00ff"`, a tuple) |
+| `trim_border` | `False` | Crop fully transparent borders away |
+| `output_format` | `"PNG"` | Encoding used when the input was bytes |
+| `quality` | `None` | JPEG/WEBP quality when the input was bytes |
+
+`Remover` is lazy and thread-safe: constructing one touches neither the network
+nor the disk, and the ONNX session is built on first use and then shared.
+
+## CLI
+
+```bash
+bgremover remove photo.jpg                        # -> photo_nobg.png
+bgremover remove photo.jpg -o out.png --refine --trim
+bgremover remove *.jpg -o cutouts/ -m isnet-general-use
+bgremover remove photo.jpg --bg white -o flat.jpg
+bgremover remove photo.jpg --only-mask -o mask.png
+cat photo.jpg | bgremover remove - > out.png
+
+bgremover models                                  # list models, show what is cached
+bgremover download u2net isnet-general-use        # pre-fetch weights
+bgremover download --all
+bgremover download --check u2net                  # re-hash a cached file
+bgremover serve --port 3001                       # run the HTTP API
+```
+
+## HTTP API
+
+```bash
+pip install "bgremover-onnx[api]"
+bgremover serve --port 3001          # docs at http://localhost:3001/docs
+```
+
+| Route | What it does |
+| --- | --- |
+| `POST /remove-background` | Multipart upload, field name `image` |
+| `POST /remove-background-url` | JSON body with `imageUrl` (or `image_url`) |
+| `GET /models` | Models this server will serve |
+| `GET /health` | Health check |
+| `GET /` | API description |
+
+```bash
+curl -X POST -F "image=@photo.jpg" \
+  http://localhost:3001/remove-background --output out.png
+
+curl -X POST -H "Content-Type: application/json" \
+  -d '{"imageUrl": "https://example.com/photo.jpg"}' \
+  http://localhost:3001/remove-background-url --output out.png
+```
+
+Both routes accept the same options as the library: `model`, `only_mask`,
+`refine`, `background`, `trim`, `format`, `quality` — as form fields on the
+upload route, as JSON keys on the URL route.
+
+Errors are always `{"error": "..."}` with a 4xx or 5xx status.
+
+### Server environment variables
+
+| Variable | Default | Meaning |
+| --- | --- | --- |
+| `BGREMOVER_MAX_UPLOAD_BYTES` | `10485760` | Upload / fetch size limit (10 MB) |
+| `BGREMOVER_MODELS` | the six models ≤ 180 MB | Comma-separated allow-list |
+| `BGREMOVER_DEFAULT_MODEL` | `u2net` | Model used when a request names none |
+| `BGREMOVER_MAX_CONCURRENCY` | `2` | Inferences allowed to run at once |
+| `BGREMOVER_FETCH_TIMEOUT` | `15` | Seconds allowed for fetching a remote image |
+| `BGREMOVER_MAX_REDIRECTS` | `3` | Redirect hops, each re-checked |
+| `BGREMOVER_ALLOW_PRIVATE_URLS` | `false` | Let URL fetches reach private addresses |
+| `BGREMOVER_CORS_ORIGINS` | `*` | Comma-separated CORS origins |
+| `BGREMOVER_PRELOAD` | `false` | Load the default model at startup |
+
+Library-wide: `BGREMOVER_HOME` (cache directory),
+`BGREMOVER_MODEL_BASE_URL` (weights mirror), `BGREMOVER_PROVIDERS`
+(onnxruntime execution providers), `BGREMOVER_NUM_THREADS`.
+
+The URL route is deliberately strict: only `http`/`https`, and the host must
+resolve to a public address — on every redirect hop too, so a public URL cannot
+bounce a request into `169.254.169.254` or your private network.
+
+## Models
+
+| Name | Size | Input | License | Notes |
+| --- | --- | --- | --- | --- |
+| `u2net` | 168 MB | 320² | Apache-2.0 | General purpose. The default. |
+| `u2netp` | 4 MB | 320² | Apache-2.0 | Small and fast, lower quality |
+| `u2net_human_seg` | 168 MB | 320² | Apache-2.0 | Fine-tuned for people |
+| `silueta` | 42 MB | 320² | Apache-2.0 | u2net pruned; near-u2net quality |
+| `isnet-general-use` | 170 MB | 1024² | Apache-2.0 | Sharper edges, slower |
+| `isnet-anime` | 168 MB | 1024² | Apache-2.0 | Anime and illustration |
+| `birefnet-general-lite` | 214 MB | 1024² | MIT | Best quality that fits in ~8 GB RAM |
+| `birefnet-general` | 928 MB | 1024² | MIT | Highest quality; needs a GPU or 16 GB+ RAM |
+
+Every entry carries a SHA-256 that is checked on download; a mismatch is
+discarded rather than cached.
+
+The two BiRefNet models are **not** in the server's default allow-list: they are
+large and memory-hungry. Add them with `BGREMOVER_MODELS` only on a machine that
+can hold them.
+
+## Docker
+
+```bash
+docker build -t bgremover .                       # bakes in u2net
+docker build --build-arg MODELS=u2netp,silueta -t bgremover .
+docker run -p 3001:3001 bgremover
+```
+
+The weights are downloaded in a separate build stage and copied into the final
+image, so a container does not fetch a model on its first request. Set
+`BGREMOVER_MODELS` at run time to match what you baked in.
+
+## Development
+
+```bash
+git clone https://github.com/asimmakhmudov/bgremover.git
+cd bgremover
+
+# with uv
+uv venv && uv pip install -e ".[dev]"
+
+# or with pip
+python -m venv .venv && source .venv/bin/activate && pip install -e ".[dev]"
+
+pytest -q                     # fast: a fake ONNX session, no network
+pytest -q -m model            # also runs real weights (downloads ~5 MB)
+ruff check src tests && ruff format --check src tests
+```
+
+Contributions are welcome. Please open an issue first for larger changes, and
+keep `pytest -q` and `ruff` green.
+
+## License
+
+MIT. The model weights carry their own licenses, listed in the table above.

bgremover_onnx-2.0.0/README.md

@@ -0,0 +1,248 @@
+# bgremover
+
+[![PyPI](https://img.shields.io/pypi/v/bgremover-onnx.svg)](https://pypi.org/project/bgremover-onnx/)
+[![Python](https://img.shields.io/pypi/pyversions/bgremover-onnx.svg)](https://pypi.org/project/bgremover-onnx/)
+[![CI](https://github.com/asimmakhmudov/bgremover/actions/workflows/ci.yml/badge.svg)](https://github.com/asimmakhmudov/bgremover/actions/workflows/ci.yml)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://github.com/asimmakhmudov/bgremover/blob/master/LICENSE)
+
+Remove image backgrounds with pretrained ONNX segmentation models — as a Python
+library, a command line tool, or an HTTP API.
+
+It runs U²-Net, IS-Net and BiRefNet with its own inference code: numpy and
+Pillow for the image math, onnxruntime for the model. There is no `rembg`
+dependency. (The weight files are hosted on rembg's releases page, which is the
+only thing taken from that project.)
+
+```python
+from bgremover import remove
+
+with open("photo.jpg", "rb") as fh:
+    png = remove(fh.read())
+```
+
+## Install
+
+With pip:
+
+```bash
+pip install bgremover-onnx            # library + CLI
+pip install "bgremover-onnx[api]"     # also the HTTP API
+```
+
+With [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add bgremover-onnx                 # add to a uv project
+uv add "bgremover-onnx[api]"
+uv pip install bgremover-onnx         # into the active environment
+uvx --from bgremover-onnx bgremover remove photo.jpg   # run the CLI without installing it
+uvx --from "bgremover-onnx[api]" bgremover serve   # run the API without installing it
+```
+
+The package is published on PyPI as `bgremover-onnx`; you still import it as
+`bgremover` and run the `bgremover` command.
+
+Python 3.10 or newer. Model weights are downloaded on first use and cached in
+`~/.cache/bgremover` (override with `BGREMOVER_HOME`).
+
+## Project structure
+
+```
+bgremover/
+├── src/bgremover/
+│   ├── __init__.py        Public API, resolved lazily (remove, Remover, MODELS, ...)
+│   ├── models.py          Model registry: URL, SHA-256, input size, normalisation
+│   ├── download.py        Download, hash-check and cache weights (stdlib only)
+│   ├── session.py         Remover: owns the ONNX session, runs the pipeline
+│   ├── processing.py      Pure image math: preprocess, postprocess, refine, cutout, trim
+│   ├── images.py          Decode any input (bytes / path / PIL / numpy), encode output
+│   ├── cli.py             `bgremover` command: remove, models, download, serve
+│   └── server/            Optional HTTP API (installed with the [api] extra)
+│       ├── app.py         FastAPI app, routes, error handling, bounded worker pool
+│       ├── config.py      Settings read from BGREMOVER_* environment variables
+│       └── fetch.py       SSRF-safe image fetching for the URL route
+├── tests/                 pytest suite (fake ONNX session; real models with -m model)
+├── .github/workflows/
+│   ├── ci.yml             Lint, test on 3.10-3.13, build the package, smoke-test Docker
+│   └── publish.yml        Publish to PyPI when a GitHub release is published
+├── Dockerfile             Two stages: bake weights in, then a slim non-root runtime
+└── pyproject.toml         Package metadata, extras, ruff and pytest config
+```
+
+How a request flows:
+
+```
+input (bytes / path / PIL / numpy)
+  -> images.load_image          decode to an RGB PIL image
+  -> processing.preprocess      resize to the model's input, normalise, NCHW tensor
+  -> onnxruntime                run the segmentation model (session.Remover)
+  -> processing.postprocess     scale the mask back to the image size
+  -> processing.refine_mask     optional: harden + feather the edge
+  -> processing.cutout / trim / apply_background
+  -> images.encode_image        same type out as came in
+```
+
+The core library never imports the `server` package, and `processing.py` does no
+I/O, so each layer can be tested on its own.
+
+## Library
+
+`remove()` returns the same type you give it: bytes in, bytes out; a PIL image
+in, a PIL image out; a numpy array in, a numpy array out; a path in, a PIL image
+out.
+
+```python
+from bgremover import remove, Remover
+
+png_bytes = remove(jpeg_bytes)                     # -> bytes (PNG)
+image     = remove("photo.jpg")                    # -> PIL.Image (RGBA)
+array     = remove(numpy_rgb)                      # -> numpy array (RGBA)
+
+# Reuse one model across many images - the session is built once.
+remover = Remover("isnet-general-use")
+for path in paths:
+    remover.remove(path).save(path.with_suffix(".png"))
+```
+
+Options for `remove()` / `Remover.remove()`:
+
+| Option | Default | Meaning |
+| --- | --- | --- |
+| `only_mask` | `False` | Return the grayscale mask instead of the cutout |
+| `refine` | `False` | Harden the confident regions and feather the edge |
+| `foreground_threshold` | `240` | With `refine`: at or above this is fully opaque |
+| `background_threshold` | `15` | With `refine`: at or below this is fully transparent |
+| `erode_size` | `0` | With `refine`: pull the edge in by N pixels |
+| `blur_radius` | `1.0` | With `refine`: feather radius |
+| `background` | `None` | Composite onto a colour (`"white"`, `"#ff00ff"`, a tuple) |
+| `trim_border` | `False` | Crop fully transparent borders away |
+| `output_format` | `"PNG"` | Encoding used when the input was bytes |
+| `quality` | `None` | JPEG/WEBP quality when the input was bytes |
+
+`Remover` is lazy and thread-safe: constructing one touches neither the network
+nor the disk, and the ONNX session is built on first use and then shared.
+
+## CLI
+
+```bash
+bgremover remove photo.jpg                        # -> photo_nobg.png
+bgremover remove photo.jpg -o out.png --refine --trim
+bgremover remove *.jpg -o cutouts/ -m isnet-general-use
+bgremover remove photo.jpg --bg white -o flat.jpg
+bgremover remove photo.jpg --only-mask -o mask.png
+cat photo.jpg | bgremover remove - > out.png
+
+bgremover models                                  # list models, show what is cached
+bgremover download u2net isnet-general-use        # pre-fetch weights
+bgremover download --all
+bgremover download --check u2net                  # re-hash a cached file
+bgremover serve --port 3001                       # run the HTTP API
+```
+
+## HTTP API
+
+```bash
+pip install "bgremover-onnx[api]"
+bgremover serve --port 3001          # docs at http://localhost:3001/docs
+```
+
+| Route | What it does |
+| --- | --- |
+| `POST /remove-background` | Multipart upload, field name `image` |
+| `POST /remove-background-url` | JSON body with `imageUrl` (or `image_url`) |
+| `GET /models` | Models this server will serve |
+| `GET /health` | Health check |
+| `GET /` | API description |
+
+```bash
+curl -X POST -F "image=@photo.jpg" \
+  http://localhost:3001/remove-background --output out.png
+
+curl -X POST -H "Content-Type: application/json" \
+  -d '{"imageUrl": "https://example.com/photo.jpg"}' \
+  http://localhost:3001/remove-background-url --output out.png
+```
+
+Both routes accept the same options as the library: `model`, `only_mask`,
+`refine`, `background`, `trim`, `format`, `quality` — as form fields on the
+upload route, as JSON keys on the URL route.
+
+Errors are always `{"error": "..."}` with a 4xx or 5xx status.
+
+### Server environment variables
+
+| Variable | Default | Meaning |
+| --- | --- | --- |
+| `BGREMOVER_MAX_UPLOAD_BYTES` | `10485760` | Upload / fetch size limit (10 MB) |
+| `BGREMOVER_MODELS` | the six models ≤ 180 MB | Comma-separated allow-list |
+| `BGREMOVER_DEFAULT_MODEL` | `u2net` | Model used when a request names none |
+| `BGREMOVER_MAX_CONCURRENCY` | `2` | Inferences allowed to run at once |
+| `BGREMOVER_FETCH_TIMEOUT` | `15` | Seconds allowed for fetching a remote image |
+| `BGREMOVER_MAX_REDIRECTS` | `3` | Redirect hops, each re-checked |
+| `BGREMOVER_ALLOW_PRIVATE_URLS` | `false` | Let URL fetches reach private addresses |
+| `BGREMOVER_CORS_ORIGINS` | `*` | Comma-separated CORS origins |
+| `BGREMOVER_PRELOAD` | `false` | Load the default model at startup |
+
+Library-wide: `BGREMOVER_HOME` (cache directory),
+`BGREMOVER_MODEL_BASE_URL` (weights mirror), `BGREMOVER_PROVIDERS`
+(onnxruntime execution providers), `BGREMOVER_NUM_THREADS`.
+
+The URL route is deliberately strict: only `http`/`https`, and the host must
+resolve to a public address — on every redirect hop too, so a public URL cannot
+bounce a request into `169.254.169.254` or your private network.
+
+## Models
+
+| Name | Size | Input | License | Notes |
+| --- | --- | --- | --- | --- |
+| `u2net` | 168 MB | 320² | Apache-2.0 | General purpose. The default. |
+| `u2netp` | 4 MB | 320² | Apache-2.0 | Small and fast, lower quality |
+| `u2net_human_seg` | 168 MB | 320² | Apache-2.0 | Fine-tuned for people |
+| `silueta` | 42 MB | 320² | Apache-2.0 | u2net pruned; near-u2net quality |
+| `isnet-general-use` | 170 MB | 1024² | Apache-2.0 | Sharper edges, slower |
+| `isnet-anime` | 168 MB | 1024² | Apache-2.0 | Anime and illustration |
+| `birefnet-general-lite` | 214 MB | 1024² | MIT | Best quality that fits in ~8 GB RAM |
+| `birefnet-general` | 928 MB | 1024² | MIT | Highest quality; needs a GPU or 16 GB+ RAM |
+
+Every entry carries a SHA-256 that is checked on download; a mismatch is
+discarded rather than cached.
+
+The two BiRefNet models are **not** in the server's default allow-list: they are
+large and memory-hungry. Add them with `BGREMOVER_MODELS` only on a machine that
+can hold them.
+
+## Docker
+
+```bash
+docker build -t bgremover .                       # bakes in u2net
+docker build --build-arg MODELS=u2netp,silueta -t bgremover .
+docker run -p 3001:3001 bgremover
+```
+
+The weights are downloaded in a separate build stage and copied into the final
+image, so a container does not fetch a model on its first request. Set
+`BGREMOVER_MODELS` at run time to match what you baked in.
+
+## Development
+
+```bash
+git clone https://github.com/asimmakhmudov/bgremover.git
+cd bgremover
+
+# with uv
+uv venv && uv pip install -e ".[dev]"
+
+# or with pip
+python -m venv .venv && source .venv/bin/activate && pip install -e ".[dev]"
+
+pytest -q                     # fast: a fake ONNX session, no network
+pytest -q -m model            # also runs real weights (downloads ~5 MB)
+ruff check src tests && ruff format --check src tests
+```
+
+Contributions are welcome. Please open an issue first for larger changes, and
+keep `pytest -q` and `ruff` green.
+
+## License
+
+MIT. The model weights carry their own licenses, listed in the table above.

bgremover_onnx-2.0.0/pyproject.toml

@@ -0,0 +1,80 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "bgremover-onnx"
+dynamic = ["version"]
+description = "Remove image backgrounds with ONNX segmentation models. Library, CLI and HTTP API."
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "MIT" }
+authors = [{ name = "Asim Mahmudov" }]
+keywords = ["background-removal", "segmentation", "onnx", "u2net", "isnet", "birefnet"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Environment :: Console",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Multimedia :: Graphics",
+    "Topic :: Scientific/Engineering :: Image Processing",
+]
+dependencies = [
+    "numpy>=1.24",
+    "pillow>=10.0",
+    "onnxruntime>=1.16",
+]
+
+[project.optional-dependencies]
+api = [
+    "fastapi>=0.110",
+    "uvicorn>=0.27",
+    "python-multipart>=0.0.9",
+]
+dev = [
+    "bgremover-onnx[api]",
+    "pytest>=8.0",
+    "httpx>=0.27",
+    "ruff>=0.5",
+    "build>=1.2",
+    "twine>=5.0",
+]
+
+[project.urls]
+Homepage = "https://github.com/asimmakhmudov/bgremover"
+Issues = "https://github.com/asimmakhmudov/bgremover/issues"
+Documentation = "https://github.com/asimmakhmudov/bgremover#readme"
+Source = "https://github.com/asimmakhmudov/bgremover"
+
+[project.scripts]
+bgremover = "bgremover.cli:main"
+
+[tool.hatch.version]
+path = "src/bgremover/__init__.py"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/bgremover"]
+
+[tool.hatch.build.targets.sdist]
+include = ["src/bgremover", "tests", "README.md", "LICENSE", "pyproject.toml"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py310"
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "UP", "B"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+# Model tests download real weights, so they are opt-in: `pytest -m model`.
+addopts = "-m 'not model'"
+markers = [
+    "model: downloads and runs a real ONNX model (needs network)",
+]
+filterwarnings = ["error::DeprecationWarning"]