# Module-Implant
This project can help you implant 3rd party / custom pytorch nn.module components
into Ultralytics library. After that, you can simply modify the yaml file,
which defines your network structure, with new components.
## 1. Install
There are 2 ways to install this project.
You can simply download / clone the project and follow the manual instructions below, or try:
```bash
pip install moduleimplant
```
## 2. Usage
**Auto setup**
Module-Implant need some manual operations to modify Ultralytics since Ultralytics does not provide APIs for custom modules and using if-else to select inner modules.
For auto setup, try:
```bash
moduleimlpant modify
```
for modifying Ulralytics pack;
```bash
moduleimlpant de-modify
```
for cancle modification before uninstallation.
**manual setup**
Paste moduleimplant.py file where task.py located, modify task.py as following:
1. import
```python
from .moduleimplant import ModuleImplant
```
2. update globals() during parse_model()
```python
globals().update(ModuleImplant.get_third_party_modules_dict())
# insert this before 'for i, (f, n, m, args) in enumerate...'
```
3. add this elif within others in the for statement that mentioned above
```python
elif m in ModuleImplant.get_third_party_modules():
c1, c2, args = ModuleImplant.parse_third_party_module(ch, f, n, m, args)
```
Then, modify the paralled `__init__.py` file as following:
1. add this piece of code
```python
from .moduleimplant import (
ModuleImplant
)
```
2. add "ModuleImplant" to `__all__`
**prepare custom modules**
You also need a parser func to parse arguments provided by .yaml file.
Implement a func called 'yaml_args_parser' in your custom module like this:
```python
@staticmethod
def yaml_args_parser(ch, f, n, m, args):
return ch[f], ch[f], [ch[f], *args]
```
tasks.py will pass 5 arguments for your custom module:
- ch: a list that recorded all layers' output channel count
- f: previous layer's serial number which indicates where this layer's data comes from
- n: assume that your layer should be a serial of modules, n represents the repetition times
- m: definition form of this module
- args: custom arguments after aboves
tasks.py needs 3 return values:
- c1: layer's input channel count
- c2: layer's output channel count
- args: all arguments that module's initial function requires
Raw data
{
"_id": null,
"home_page": "https://github.com/Gelinzh/moduleimplant",
"name": "moduleimplant",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": null,
"author": "acelych",
"author_email": "acelych@foxmail.com",
"download_url": "https://files.pythonhosted.org/packages/e4/35/bce3cf8831179a9aaf6c90178c221e56118ba0eb88a0b29e5f1c2213fed5/moduleimplant-0.2.0.tar.gz",
"platform": null,
"description": "# Module-Implant\n\nThis project can help you implant 3rd party / custom pytorch nn.module components\ninto Ultralytics library. After that, you can simply modify the yaml file,\nwhich defines your network structure, with new components.\n\n## 1. Install\n\nThere are 2 ways to install this project.\nYou can simply download / clone the project and follow the manual instructions below, or try:\n\n```bash\npip install moduleimplant\n```\n\n## 2. Usage\n\n**Auto setup**\n\nModule-Implant need some manual operations to modify Ultralytics since Ultralytics does not provide APIs for custom modules and using if-else to select inner modules.\n\nFor auto setup, try:\n\n```bash\nmoduleimlpant modify\n```\n\nfor modifying Ulralytics pack;\n\n```bash\nmoduleimlpant de-modify\n```\n\nfor cancle modification before uninstallation.\n\n**manual setup**\n\nPaste moduleimplant.py file where task.py located, modify task.py as following:\n1. import\n ```python\n from .moduleimplant import ModuleImplant\n ```\n2. update globals() during parse_model()\n ```python\n globals().update(ModuleImplant.get_third_party_modules_dict())\n # insert this before 'for i, (f, n, m, args) in enumerate...'\n ```\n3. add this elif within others in the for statement that mentioned above\n ```python\n elif m in ModuleImplant.get_third_party_modules():\n c1, c2, args = ModuleImplant.parse_third_party_module(ch, f, n, m, args)\n ```\n \nThen, modify the paralled `__init__.py` file as following:\n1. add this piece of code\n ```python\n from .moduleimplant import (\n ModuleImplant\n )\n ```\n2. add \"ModuleImplant\" to `__all__`\n\n**prepare custom modules**\n\nYou also need a parser func to parse arguments provided by .yaml file.\nImplement a func called 'yaml_args_parser' in your custom module like this:\n```python\n@staticmethod\ndef yaml_args_parser(ch, f, n, m, args):\n return ch[f], ch[f], [ch[f], *args]\n```\ntasks.py will pass 5 arguments for your custom module:\n- ch: a list that recorded all layers' output channel count\n- f: previous layer's serial number which indicates where this layer's data comes from\n- n: assume that your layer should be a serial of modules, n represents the repetition times\n- m: definition form of this module\n- args: custom arguments after aboves\n\ntasks.py needs 3 return values:\n- c1: layer's input channel count\n- c2: layer's output channel count\n- args: all arguments that module's initial function requires\n",
"bugtrack_url": null,
"license": null,
"summary": "A custom module implant system for Ultralytics models",
"version": "0.2.0",
"project_urls": {
"Homepage": "https://github.com/Gelinzh/moduleimplant"
},
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "18fa1145b7a8f945ad710bf05281dd19921c1e4ebcfeda32291994b2f7a8096b",
"md5": "1de8cceecdb654ea6a4064996c622a23",
"sha256": "8d76de72338fd37168be9b7039f50231c0971019121fcba64a99f6881a391c35"
},
"downloads": -1,
"filename": "moduleimplant-0.2.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "1de8cceecdb654ea6a4064996c622a23",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.9",
"size": 7007,
"upload_time": "2024-12-21T16:14:46",
"upload_time_iso_8601": "2024-12-21T16:14:46.988904Z",
"url": "https://files.pythonhosted.org/packages/18/fa/1145b7a8f945ad710bf05281dd19921c1e4ebcfeda32291994b2f7a8096b/moduleimplant-0.2.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "e435bce3cf8831179a9aaf6c90178c221e56118ba0eb88a0b29e5f1c2213fed5",
"md5": "78f805893d41dcb6d1cfc36c8c500980",
"sha256": "6383cdb1b2bd052e592a296bac82afce4b4357034e4dfba97693892b36bd1658"
},
"downloads": -1,
"filename": "moduleimplant-0.2.0.tar.gz",
"has_sig": false,
"md5_digest": "78f805893d41dcb6d1cfc36c8c500980",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.9",
"size": 5579,
"upload_time": "2024-12-21T16:15:00",
"upload_time_iso_8601": "2024-12-21T16:15:00.301501Z",
"url": "https://files.pythonhosted.org/packages/e4/35/bce3cf8831179a9aaf6c90178c221e56118ba0eb88a0b29e5f1c2213fed5/moduleimplant-0.2.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-12-21 16:15:00",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "Gelinzh",
"github_project": "moduleimplant",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "moduleimplant"
}
JSON
