Metadata-Version: 2.4
Name: easy-image-model
Version: 0.1.2
Summary: Easy PyTorch image classification library
Author-email: Harrison Hensarling <harrisonhensarling@gmail.com>
Maintainer-email: Harrison Hensarling <harrisonhensarling@gmail.com>
License-Expression: MIT
Project-URL: Homepage, https://github.com/HHensar/easy_image_model
Project-URL: Documentation, https://github.com/HHensar/easy_image_model#readme
Project-URL: Repository, https://github.com/HHensar/easy_image_model
Project-URL: Issues, https://github.com/HHensar/easy_image_model/issues
Project-URL: Changelog, https://github.com/HHensar/easy_image_model/blob/main/CHANGELOG.md
Keywords: machine learning,pytorch,image classification
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Operating System :: OS Independent
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
Requires-Dist: torch>=2.0
Requires-Dist: torchvision>=0.15
Requires-Dist: pillow>=9.0
Provides-Extra: dev
Requires-Dist: pytest>=7.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0; extra == "dev"
Requires-Dist: black>=22.0; extra == "dev"
Requires-Dist: isort>=5.0; extra == "dev"
Requires-Dist: flake8>=4.0; extra == "dev"
Requires-Dist: mypy>=1.0; extra == "dev"

# Easy Image Model
### by Harrison Hensarling
**Easy Image Model** is a lightweight Python library for quickly building, training, and evaluating image classification models using PyTorch. 
It is designed for simplicity and ease of use. Supports:

-**small datasets**

-**non-normalized images**

-**batch training**  

---

## Features

- Create a custom neural network with **any number of hidden layers and nodes**.  
- Train models on **folders of labeled images** (each folder represents a category).  
- Batch training for faster convergence.  
- Automatically preprocesses images (resizing, normalization) — **no manual preprocessing required except for labeling**.  
- Evaluate a single image and return **category probabilities**.  
- Save and reload model weights as JSON for portability.  
- **Customizable image input size** (default 224x224, can be adjusted for your dataset).
- Supported image types: 
   - `.jpg` 
   - `.jpeg`
   - `.png`
   - `.bmp`
   - `.gif (only first frame is used)`
   - `.tif`
   - `.tiff`
   - ` More supported but untested`
---

## Installation

```bash
pip install easy-image-model
```

Or for development:

```bash
git clone https://github.com/hhensar/easy-image-model.git
cd easy-image-model
pip install -e ".[dev]"
```

---

## Example Usage

```python
from easy_image_model import create_model, train_model_batch_folders, evaluate_model

# Define categories
categories = ['Eagles', 'Penguins', 'Owls', 'Others']

# Create model with hidden layers (3 layers with 512, 256, 128 nodes)
# img_size defaults to 224, but can be customized (e.g., 128, 256, 512)
model = create_model([512, 256, 128], categories, img_size=224)

# Train on folders (each folder contains images of one category)
folder_paths = {
    'Eagles': 'dataset/Eagles',
    'Penguins': 'dataset/Penguins',
    'Owls': 'dataset/Owls',
    'Others': 'dataset/Others'
}
model = train_model_batch_folders(model, folder_paths, batch_size=4, epochs=5)

# Evaluate a single image
result = evaluate_model(model, 'test_images/test1.jpg')
print(result)
```

**Example Output:**
```
{'Eagles': 0.87, 
 'Penguins': 0.05, 
 'Owls': 0.23, 
 'Others': 0.1}
```

# Documentation

## Functions

### `create_model(layers, categories, img_size=224, channels=3)`

Creates a new image classification model.

#### Parameters
- **`layers`** (`list[int]`)  
  List of positive integers specifying the hidden layer sizes.

- **`categories`** (`list[str]`)  
  List of unique category names for classification.

- **`img_size`** (`int`, optional)  
  Image size (default: `224`). Images are resized to `img_size × img_size`.

- **`channels`** (`int`, optional)  
  Number of image channels (default: `3`).

#### Returns
- **`dict`**  
  Model configuration object used for training and evaluation.

#### Raises
- `ValueError` if layers or categories are invalid.

---

### `train_model(model, img_path, labels, lr=1e-4)`

Trains the model on a single image with the provided labels.

#### Parameters
- **`model`** (`dict`)  
  Model configuration returned by `create_model`.

- **`img_path`** (`str`)  
  Path to the image file.

- **`labels`** (`list[str]`)  
  List of category labels associated with the image.

- **`lr`** (`float`, optional)  
  Learning rate (default: `1e-4`).

#### Returns
- **`dict`**  
  Updated model configuration after training.

#### Raises
- `ValueError` if the model is invalid.
- `FileNotFoundError` if `img_path` does not exist.

---

### `train_model_batch_folders(model, folder_paths, batch_size=4, epochs=5, lr=1e-4)`

Trains the model using batches of images from folders, where each folder represents a category.

#### Parameters
- **`model`** (`dict`)  
  Model configuration returned by `create_model`.

- **`folder_paths`** (`dict[str, str]`)  
  Mapping of category names to folder paths containing images.

- **`batch_size`** (`int`, optional)  
  Number of images per batch (default: `4`).

- **`epochs`** (`int`, optional)  
  Number of training epochs (default: `5`).

- **`lr`** (`float`, optional)  
  Learning rate (default: `1e-4`).

#### Returns
- **`dict`**  
  Updated model configuration after training.

#### Raises
- `ValueError` if inputs are invalid.
- `FileNotFoundError` if any folder paths do not exist.

---

### `evaluate_model(model, img_path)`

Evaluates the model on a single image and returns classification probabilities.

#### Parameters
- **`model`** (`dict`)  
  Trained model configuration.

- **`img_path`** (`str`)  
  Path to the image file.

#### Returns
- **`dict[str, float]`**  
  Mapping of category names to probability scores.

#### Raises
- `ValueError` if the model is invalid.
- `FileNotFoundError` if `img_path` does not exist.


## Example Usage

The following example shows how to create a model, train it using folders of labeled images, and evaluate a new image.

### Folder Structure

Each category should have its own folder containing images:

```

dataset/
├── Eagles/
│   ├── img1.jpg
│   └── img2.jpg
├── Penguins/
├── Owls/
└── Others/

````

### Training and Evaluation

```python
from easy_image_model import (
    create_model,
    train_model_batch_folders,
    evaluate_model
)

# Define classification categories
categories = ['Eagles', 'Penguins', 'Owls', 'Others']

# Create a model with three hidden layers
model = create_model(
    layers=[512, 256, 128],
    categories=categories,
    img_size=224
)

# Map categories to their image folders
folder_paths = {
    'Eagles': 'dataset/Eagles',
    'Penguins': 'dataset/Penguins',
    'Owls': 'dataset/Owls',
    'Others': 'dataset/Others'
}

# Train the model
model = train_model_batch_folders(
    model,
    folder_paths,
    batch_size=4,
    epochs=5
)

# Evaluate a new image
result = evaluate_model(model, 'test_images/test1.jpg')

print(result)
# Example output:
# {
#   'Eagles': 0.87,
#   'Penguins': 0.05,
#   'Owls': 0.03,
#   'Others': 0.05
# }
````

The returned dictionary maps each category to its predicted probability for the input image.
