seavision-python 0.1.1__tar.gz

seavision_python-0.1.1/PKG-INFO

@@ -0,0 +1,286 @@
+Metadata-Version: 2.4
+Name: seavision-python
+Version: 0.1.1
+Summary: Human-in-the-loop Computer Vision for Marine Environments
+Author: Mario Lambrette
+License: Proprietary
+Project-URL: Homepage, https://github.com/mariolambrette/SeaVision
+Project-URL: Documentation, https://github.com/mariolambrette/SeaVision/tree/Main/docs
+Project-URL: Source, https://github.com/mariolambrette/SeaVision
+Project-URL: Issues, https://github.com/mariolambrette/SeaVision/issues
+Keywords: computer-vision,detection,marine,SAM3
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: numpy<2.0,>=1.21
+Requires-Dist: opencv-python-headless>=4.8
+Requires-Dist: PyYAML>=6.0
+Requires-Dist: tqdm>=4.60
+Provides-Extra: s3
+Requires-Dist: boto3>=1.26; extra == "s3"
+Provides-Extra: gui
+Requires-Dist: boto3>=1.26; extra == "gui"
+Requires-Dist: PySide6<6.9,>=6.5; extra == "gui"
+Requires-Dist: psutil>=5.9; extra == "gui"
+Provides-Extra: edge
+Requires-Dist: onnxruntime>=1.16; extra == "edge"
+Provides-Extra: export
+Requires-Dist: onnxruntime>=1.16; extra == "export"
+Requires-Dist: ultralytics>=8.3.237; extra == "export"
+Provides-Extra: sam3
+Requires-Dist: pillow>=10.0; extra == "sam3"
+Requires-Dist: torch>=2.0; extra == "sam3"
+Requires-Dist: torchvision>=0.20; extra == "sam3"
+Requires-Dist: transformers>=4.47; extra == "sam3"
+Requires-Dist: ultralytics>=8.3.237; extra == "sam3"
+Provides-Extra: all
+Requires-Dist: boto3>=1.26; extra == "all"
+Requires-Dist: onnxruntime>=1.16; extra == "all"
+Requires-Dist: pillow>=10.0; extra == "all"
+Requires-Dist: psutil>=5.9; extra == "all"
+Requires-Dist: PySide6<6.9,>=6.5; extra == "all"
+Requires-Dist: torch>=2.0; extra == "all"
+Requires-Dist: torchvision>=0.20; extra == "all"
+Requires-Dist: transformers>=4.47; extra == "all"
+Requires-Dist: ultralytics>=8.3.237; extra == "all"
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == "dev"
+Requires-Dist: pytest>=8.0; extra == "dev"
+Requires-Dist: pytest-cov>=5.0; extra == "dev"
+Requires-Dist: twine>=5.1; extra == "dev"
+
+<p align="left">
+  <img src="https://raw.githubusercontent.com/mariolambrette/SeaVision/Main/assets/seavision_logo.jpg" alt="Logo" width="200"/>
+</p>
+
+# SeaVision
+
+## Human-in-the-loop Computer Vision for Marine Environments
+
+> [!WARNING]
+> Documentation is currently out of date.
+> The packaging, release, and edge-deployment flows have changed recently, and
+> some instructions in this README and the linked docs may still describe older
+> clone-first or pre-release workflows. Treat the published release artifacts
+> and current CLI behavior as the source of truth until the docs are refreshed.
+
+SeaVision offers a practical computer vision workflow for marine monitoring. It supports the development of accurate, 
+custom detection models for specific deployment scenarios. The main workflow is:
+
+1. Deploy a generic detector (e.g. motion detection, SAM3-based semantic detection, etc.) on recorded footage.
+2. Use the provided GUI to validate and refine detections into a training dataset.
+3. Train a custom YOLO detection model.
+4. Deploy the trained model on a desktop workstation or edge device with the provided export functionality.
+
+If you are new to the project, start with the [Start Here guide](./docs/Start_Here.md).
+
+## Start Here (New Users)
+
+If you are not a computer vision specialist, use this short path first:
+
+1. Read the [Start Here guide](./docs/Start_Here.md) for a plain-language workflow and glossary.
+2. Install SeaVision in GUI mode: `python -m pip install ".[gui]"`.
+3. Run `seavision-gui` and create a new session with `Ctrl+N`.
+4. Follow the GUI [Getting Started guide](./docs/gui/Getting%20started.md).
+
+## Installation
+
+### Prerequisites
+
+- Python 3.12
+- pip
+
+### Install from a cloned repository
+
+Before installing the package, you must currently clone this repository. Run the following in a
+bash terminal:
+
+```bash
+git clone https://github.com/mariolambrette/SeaVision
+cd SeaVision
+```
+
+It is recommended to install SeaVision in a dedicated environment. If you are a [conda](https://www.anaconda.com/docs/getting-started/anaconda/install/overview) user, run
+the following:
+
+```bash
+conda create seavision
+conda activate seavision
+```
+
+You can now install SeaVision using pip:
+
+```bash
+pip install ".[<extras>]"
+```
+
+SeaVision can be installed in various modes, each of which supports different modes of functionality. See below for a full 
+description of all installation modes.
+
+### Install Matrix (User Types)
+
+Use the install command that matches your use case.
+
+| User type | Install command | Primary CLI command(s) |
+|---|---|---|
+| Detection-only users (local videos) | `python -m pip install .` | `seavision` |
+| Detection users with videos in S3 bucket | `python -m pip install ".[s3]"` | `seavision` |
+| GUI users (includes S3 support) | `python -m pip install ".[gui]"` | `seavision-gui` |
+| Edge device operators (runtime only) | `python -m pip install ".[edge]"` | `seavision-edge` |
+| Export users (prepare edge artifacts) | `python -m pip install ".[export]"` | `seavision-export` |
+| Advanced/full users | `python -m pip install ".[all]"` | `seavision`, `seavision-gui`, `seavision-export`, `seavision-edge` |
+
+Most users should use either `python -m pip install ".[gui]"` for desktop-only usage or `python -m pip install ".[all]"` if they are
+also likely to use edge deployment functionality. Other installation methods may provide lighter dependencies in specific scenarios.
+
+You can verify your installation by running [test commands](#install-verification)
+
+### Development install
+
+```bash
+python -m venv venv
+venv\Scripts\activate.bat
+python -m pip install -e ".[dev]"
+```
+
+For release/distribution policy and compatibility targets, see:
+
+- [Distribution and compatibility policy](./docs/release/Distribution_and_Compatibility.md)
+
+### Optional GPU acceleration
+
+The above install allows you to run all SeaVision functionality but does not
+provide support for GPU-accelerated processing. Some methods (e.g. SAM3-based
+detection) will be significantly faster if a GPU is available. You will need to
+install GPU-specific dependencies manually based on your hardware.
+
+For example, to run on an NVIDIA RTX 5090 GPU with cuda 13 you could install
+the following:
+
+```bash
+pip install torch>=2.9.0 torchvision>=0.20.0 --index-url https://download.pytorch.org/whl/cu130 --force-reinstall
+```
+You can find more information on PyTorch compatibility
+[here](https://pytorch.org/get-started/locally/).
+
+### AWS S3 Access (Optional)
+
+The pipeline supports streaming video directly from AWS. In order to access this
+feature you will need to configure an AWS SSO profile. For more information on
+how to do this and integrate AWS streaming into the SeaVision workflow, see the
+[AWS setup documentation](./docs/AWS_SETUP.md)
+
+## Install verification
+
+Run the command that matches your installed mode:
+
+```bash
+# Detection CLI
+seavision --help
+
+# GUI
+seavision-gui
+
+# Export CLI
+seavision-export --help
+
+# Edge CLI
+seavision-edge --help
+```
+
+## Quick start
+
+### 1. Run the detection pipeline
+
+The full detection pipeline can be run with a CLI command, which can optionally be configured
+with a [YAML file](./config/default.yaml)
+
+```bash
+# Local files
+seavision --input ./data/footage/ --output ./results/
+
+# With config file
+seavision --config config/default.yaml
+```
+
+### 2. Launch the GUI
+
+```bash
+seavision-gui
+```
+
+### 3. Export for edge deployment
+
+```bash
+seavision-export --weights models/fish.pt --output ./deployment --target onnx
+```
+
+### 4. Run on an edge device
+
+```bash
+seavision-edge --artifact-dir ./deployment/artifact
+```
+
+## Troubleshooting quick checks
+
+| Problem | Likely cause | Quick check | Fix |
+|---|---|---|---|
+| `seavision: command not found` | Package not installed in active environment | `python -m pip show seavision` | Activate the right environment and reinstall using the install matrix |
+| `seavision-gui: command not found` | GUI extra not installed | `python -m pip show PySide6` | `python -m pip install ".[gui]"` |
+| `seavision-edge: command not found` | Edge extra not installed | `python -m pip show onnxruntime` | `python -m pip install ".[edge]"` |
+| Import error for `boto3` when using S3 | S3 dependency missing | `python -m pip show boto3` | `python -m pip install ".[s3]"` or `python -m pip install ".[gui]"` |
+| GUI starts but S3 login fails | AWS credentials/session not configured | `aws sts get-caller-identity` | Run `aws sso login --profile <profile-name>` and retry |
+| Edge run fails at startup validation | Artifact dir incomplete or invalid | `seavision-edge --artifact-dir <path>` output | Re-run `seavision-export` and copy the full artifact directory again |
+
+## Configuration
+
+The SeaVision pipeline can be fully customised using YAML config files. For
+a documented example of a config file see the
+[default configuration](./config/default.yaml)
+
+
+## Edge Deployment
+
+Users may wish to deploy models developed or refined using SeaVision to an edge
+device. Full support is provided for deploying trained model weight files
+(`*.pt`) on Raspberry Pi.
+
+Edge deployment creates SeaVision detection files directly on the Pi which may
+save compute time, power and data transfer requirements. Users can optionally
+export periodic validation clips alongside the detections. See below for 
+detailed documentation:
+
+If you are new to the deployment workflow, use this order:
+
+1. [Edge deployment overview](./docs/edge/README.md)
+2. [Concepts and Terms](./docs/edge/Concepts%20and%20Terms.md)
+3. [Prerequisites Checklist](./docs/edge/Prerequisites%20Checklist.md)
+4. [Operator Quickstart](./docs/edge/Operator%20Quickstart.md)
+5. [Troubleshooting](./docs/edge/Troubleshooting.md)
+
+Use the technical references only when you are preparing the deployment files
+or troubleshooting the package structure:
+
+- [Deploy to Raspberry Pi](./docs/edge/Deploy%20to%20Raspberry%20Pi.md)
+- [Troubleshooting](./docs/edge/Troubleshooting.md)
+- [Wheel runtime workflow](./docs/edge/Wheel%20Runtime.md)
+- [Model artifact layout](./docs/edge/Model%20Artifacts.md)
+- [Rollback runbook](./docs/edge/Rollback%20Runbook.md)
+
+## Documentation
+
+- [Start Here guide](./docs/Start_Here.md)
+- [Class and API reference](./docs/Class_reference.md)
+- [Architecture and class diagrams](./docs/Class_diagram.md)
+- [AWS setup guide](./docs/AWS_SETUP.md)
+- [Edge deployment docs](./docs/edge/README.md)
+- [GUI docs](./docs/gui/README.md)
+
+## Project Status
+
+Note that this project is under active development and features may not yet
+work as intended.

seavision_python-0.1.1/README.md

@@ -0,0 +1,231 @@
+<p align="left">
+  <img src="https://raw.githubusercontent.com/mariolambrette/SeaVision/Main/assets/seavision_logo.jpg" alt="Logo" width="200"/>
+</p>
+
+# SeaVision
+
+## Human-in-the-loop Computer Vision for Marine Environments
+
+> [!WARNING]
+> Documentation is currently out of date.
+> The packaging, release, and edge-deployment flows have changed recently, and
+> some instructions in this README and the linked docs may still describe older
+> clone-first or pre-release workflows. Treat the published release artifacts
+> and current CLI behavior as the source of truth until the docs are refreshed.
+
+SeaVision offers a practical computer vision workflow for marine monitoring. It supports the development of accurate, 
+custom detection models for specific deployment scenarios. The main workflow is:
+
+1. Deploy a generic detector (e.g. motion detection, SAM3-based semantic detection, etc.) on recorded footage.
+2. Use the provided GUI to validate and refine detections into a training dataset.
+3. Train a custom YOLO detection model.
+4. Deploy the trained model on a desktop workstation or edge device with the provided export functionality.
+
+If you are new to the project, start with the [Start Here guide](./docs/Start_Here.md).
+
+## Start Here (New Users)
+
+If you are not a computer vision specialist, use this short path first:
+
+1. Read the [Start Here guide](./docs/Start_Here.md) for a plain-language workflow and glossary.
+2. Install SeaVision in GUI mode: `python -m pip install ".[gui]"`.
+3. Run `seavision-gui` and create a new session with `Ctrl+N`.
+4. Follow the GUI [Getting Started guide](./docs/gui/Getting%20started.md).
+
+## Installation
+
+### Prerequisites
+
+- Python 3.12
+- pip
+
+### Install from a cloned repository
+
+Before installing the package, you must currently clone this repository. Run the following in a
+bash terminal:
+
+```bash
+git clone https://github.com/mariolambrette/SeaVision
+cd SeaVision
+```
+
+It is recommended to install SeaVision in a dedicated environment. If you are a [conda](https://www.anaconda.com/docs/getting-started/anaconda/install/overview) user, run
+the following:
+
+```bash
+conda create seavision
+conda activate seavision
+```
+
+You can now install SeaVision using pip:
+
+```bash
+pip install ".[<extras>]"
+```
+
+SeaVision can be installed in various modes, each of which supports different modes of functionality. See below for a full 
+description of all installation modes.
+
+### Install Matrix (User Types)
+
+Use the install command that matches your use case.
+
+| User type | Install command | Primary CLI command(s) |
+|---|---|---|
+| Detection-only users (local videos) | `python -m pip install .` | `seavision` |
+| Detection users with videos in S3 bucket | `python -m pip install ".[s3]"` | `seavision` |
+| GUI users (includes S3 support) | `python -m pip install ".[gui]"` | `seavision-gui` |
+| Edge device operators (runtime only) | `python -m pip install ".[edge]"` | `seavision-edge` |
+| Export users (prepare edge artifacts) | `python -m pip install ".[export]"` | `seavision-export` |
+| Advanced/full users | `python -m pip install ".[all]"` | `seavision`, `seavision-gui`, `seavision-export`, `seavision-edge` |
+
+Most users should use either `python -m pip install ".[gui]"` for desktop-only usage or `python -m pip install ".[all]"` if they are
+also likely to use edge deployment functionality. Other installation methods may provide lighter dependencies in specific scenarios.
+
+You can verify your installation by running [test commands](#install-verification)
+
+### Development install
+
+```bash
+python -m venv venv
+venv\Scripts\activate.bat
+python -m pip install -e ".[dev]"
+```
+
+For release/distribution policy and compatibility targets, see:
+
+- [Distribution and compatibility policy](./docs/release/Distribution_and_Compatibility.md)
+
+### Optional GPU acceleration
+
+The above install allows you to run all SeaVision functionality but does not
+provide support for GPU-accelerated processing. Some methods (e.g. SAM3-based
+detection) will be significantly faster if a GPU is available. You will need to
+install GPU-specific dependencies manually based on your hardware.
+
+For example, to run on an NVIDIA RTX 5090 GPU with cuda 13 you could install
+the following:
+
+```bash
+pip install torch>=2.9.0 torchvision>=0.20.0 --index-url https://download.pytorch.org/whl/cu130 --force-reinstall
+```
+You can find more information on PyTorch compatibility
+[here](https://pytorch.org/get-started/locally/).
+
+### AWS S3 Access (Optional)
+
+The pipeline supports streaming video directly from AWS. In order to access this
+feature you will need to configure an AWS SSO profile. For more information on
+how to do this and integrate AWS streaming into the SeaVision workflow, see the
+[AWS setup documentation](./docs/AWS_SETUP.md)
+
+## Install verification
+
+Run the command that matches your installed mode:
+
+```bash
+# Detection CLI
+seavision --help
+
+# GUI
+seavision-gui
+
+# Export CLI
+seavision-export --help
+
+# Edge CLI
+seavision-edge --help
+```
+
+## Quick start
+
+### 1. Run the detection pipeline
+
+The full detection pipeline can be run with a CLI command, which can optionally be configured
+with a [YAML file](./config/default.yaml)
+
+```bash
+# Local files
+seavision --input ./data/footage/ --output ./results/
+
+# With config file
+seavision --config config/default.yaml
+```
+
+### 2. Launch the GUI
+
+```bash
+seavision-gui
+```
+
+### 3. Export for edge deployment
+
+```bash
+seavision-export --weights models/fish.pt --output ./deployment --target onnx
+```
+
+### 4. Run on an edge device
+
+```bash
+seavision-edge --artifact-dir ./deployment/artifact
+```
+
+## Troubleshooting quick checks
+
+| Problem | Likely cause | Quick check | Fix |
+|---|---|---|---|
+| `seavision: command not found` | Package not installed in active environment | `python -m pip show seavision` | Activate the right environment and reinstall using the install matrix |
+| `seavision-gui: command not found` | GUI extra not installed | `python -m pip show PySide6` | `python -m pip install ".[gui]"` |
+| `seavision-edge: command not found` | Edge extra not installed | `python -m pip show onnxruntime` | `python -m pip install ".[edge]"` |
+| Import error for `boto3` when using S3 | S3 dependency missing | `python -m pip show boto3` | `python -m pip install ".[s3]"` or `python -m pip install ".[gui]"` |
+| GUI starts but S3 login fails | AWS credentials/session not configured | `aws sts get-caller-identity` | Run `aws sso login --profile <profile-name>` and retry |
+| Edge run fails at startup validation | Artifact dir incomplete or invalid | `seavision-edge --artifact-dir <path>` output | Re-run `seavision-export` and copy the full artifact directory again |
+
+## Configuration
+
+The SeaVision pipeline can be fully customised using YAML config files. For
+a documented example of a config file see the
+[default configuration](./config/default.yaml)
+
+
+## Edge Deployment
+
+Users may wish to deploy models developed or refined using SeaVision to an edge
+device. Full support is provided for deploying trained model weight files
+(`*.pt`) on Raspberry Pi.
+
+Edge deployment creates SeaVision detection files directly on the Pi which may
+save compute time, power and data transfer requirements. Users can optionally
+export periodic validation clips alongside the detections. See below for 
+detailed documentation:
+
+If you are new to the deployment workflow, use this order:
+
+1. [Edge deployment overview](./docs/edge/README.md)
+2. [Concepts and Terms](./docs/edge/Concepts%20and%20Terms.md)
+3. [Prerequisites Checklist](./docs/edge/Prerequisites%20Checklist.md)
+4. [Operator Quickstart](./docs/edge/Operator%20Quickstart.md)
+5. [Troubleshooting](./docs/edge/Troubleshooting.md)
+
+Use the technical references only when you are preparing the deployment files
+or troubleshooting the package structure:
+
+- [Deploy to Raspberry Pi](./docs/edge/Deploy%20to%20Raspberry%20Pi.md)
+- [Troubleshooting](./docs/edge/Troubleshooting.md)
+- [Wheel runtime workflow](./docs/edge/Wheel%20Runtime.md)
+- [Model artifact layout](./docs/edge/Model%20Artifacts.md)
+- [Rollback runbook](./docs/edge/Rollback%20Runbook.md)
+
+## Documentation
+
+- [Start Here guide](./docs/Start_Here.md)
+- [Class and API reference](./docs/Class_reference.md)
+- [Architecture and class diagrams](./docs/Class_diagram.md)
+- [AWS setup guide](./docs/AWS_SETUP.md)
+- [Edge deployment docs](./docs/edge/README.md)
+- [GUI docs](./docs/gui/README.md)
+
+## Project Status
+
+Note that this project is under active development and features may not yet
+work as intended.

seavision_python-0.1.1/pyproject.toml

@@ -0,0 +1,94 @@
+[build-system]
+requires = ["setuptools>=61", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "seavision-python"
+version = "0.1.1"
+description = "Human-in-the-loop Computer Vision for Marine Environments"
+readme = "README.md"
+requires-python = ">=3.10"
+authors = [{ name = "Mario Lambrette" }]
+license = { text = "Proprietary" }
+keywords = ["computer-vision", "detection", "marine", "SAM3"]
+classifiers = [
+  "Programming Language :: Python",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+]
+dependencies = [
+  "numpy>=1.21,<2.0",
+  "opencv-python-headless>=4.8",
+  "PyYAML>=6.0",
+  "tqdm>=4.60",
+]
+
+[project.urls]
+Homepage = "https://github.com/mariolambrette/SeaVision"
+Documentation = "https://github.com/mariolambrette/SeaVision/tree/Main/docs"
+Source = "https://github.com/mariolambrette/SeaVision"
+Issues = "https://github.com/mariolambrette/SeaVision/issues"
+
+[project.optional-dependencies]
+# Optional S3 support for pipeline and GUI remote workflows.
+s3 = ["boto3>=1.26"]
+
+# GUI mode dependencies.
+gui = [
+  "boto3>=1.26",
+  "PySide6>=6.5,<6.9",
+  "psutil>=5.9",
+]
+
+# Edge runtime dependencies.
+edge = ["onnxruntime>=1.16"]
+
+# Land-side model export dependencies.
+export = [
+  "onnxruntime>=1.16",
+  "ultralytics>=8.3.237",
+]
+
+# SAM3 detector dependencies.
+sam3 = [
+  "pillow>=10.0",
+  "torch>=2.0",
+  "torchvision>=0.20",
+  "transformers>=4.47",
+  "ultralytics>=8.3.237",
+]
+
+# Combined install for workstations needing all SeaVision features.
+all = [
+  "boto3>=1.26",
+  "onnxruntime>=1.16",
+  "pillow>=10.0",
+  "psutil>=5.9",
+  "PySide6>=6.5,<6.9",
+  "torch>=2.0",
+  "torchvision>=0.20",
+  "transformers>=4.47",
+  "ultralytics>=8.3.237",
+]
+
+# For development in this repo; install with: pip install -e .[dev]
+dev = [
+  "build>=1.2",
+  "pytest>=8.0",
+  "pytest-cov>=5.0",
+  "twine>=5.1",
+]
+
+[project.scripts]
+seavision = "seavision.run_pipeline:main"
+seavision-gui = "seavision.gui.app:main"
+seavision-export = "seavision.run_export:main"
+seavision-edge = "seavision.run_edge:main"
+
+[tool.setuptools.packages.find]
+include = ["seavision*"]
+
+[tool.setuptools.package-data]
+seavision = ["resources/*.yaml"]

seavision_python-0.1.1/seavision/__init__.py

@@ -0,0 +1,3 @@
+"""SeaVision - Human-in-the-loop Computer Vision for Marine Environments."""
+
+__version__ = "0.1.1"

seavision_python-0.1.1/seavision/defaults.py

@@ -0,0 +1,38 @@
+"""Helpers for packaged default configuration files."""
+
+from importlib.resources import files
+from pathlib import Path
+
+_DEFAULT_PIPELINE_CONFIG = files("seavision").joinpath("resources/default.yaml")
+
+def default_pipeline_config_text() -> str:
+    """Return the packaged default pipeline configuration as text."""
+    return _DEFAULT_PIPELINE_CONFIG.read_text(encoding="utf-8")
+
+def write_default_pipeline_config(
+    destination: str | Path,
+    overwrite: bool = False,
+) -> Path:
+    """
+    Write the packaged default pipeline configuration to a user-selected path.
+
+    Args:
+        destination: Output path for the YAML file.
+        overwrite: If True, replace an existing file.
+
+    Returns:
+        The resolved output path.
+
+    Raises:
+        FileExistsError: If the file already exists and overwrite is False.
+    """
+    output_path = Path(destination)
+
+    if output_path.exists() and not overwrite:
+        raise FileExistsError(
+            f"Refusing to overwrite existing file: {output_path}"
+        )
+
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    output_path.write_text(default_pipeline_config_text(), encoding="utf-8")
+    return output_path.resolve()

seavision_python-0.1.1/seavision/edge/__init__.py

@@ -0,0 +1,38 @@
+"""
+SeaVision edge inference runtime.
+
+Lightweight inference package for edge devices (Raspberry Pi, Jetson,
+etc.). Runs ONNX models with minimal dependencies — no PyTorch, no
+Ultralytics.
+
+Public API:
+- EdgeConfig: Runtime configuration (always importable).
+- EdgeRuntime: Main inference loop (requires onnxruntime at access time).
+- FrameCapture: Video/camera input (requires opencv).
+- EdgeDetectionWriter: CSV output (no special dependencies).
+"""
+
+from __future__ import annotations
+from typing import TYPE_CHECKING
+
+from .config import EdgeConfig
+
+if TYPE_CHECKING:
+    # Only imported during type checking, not at runtime.
+    # Keeps onnxruntime optional on the land-side tooling.
+    from .runtime import EdgeRuntime
+
+# EdgeRuntime is imported lazily because it depends on onnxruntime,
+# which is only installed on the edge device. This allows the land-side
+# export tooling to import seavision.edge (for EdgeConfig and
+# artifact builders) without requiring onnxruntime.
+
+
+def __getattr__(name: str):
+    if name == "EdgeRuntime":
+        from .runtime import EdgeRuntime
+        return EdgeRuntime
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
+__all__ = ["EdgeConfig", "EdgeRuntime"]