# convolutional-autoencoder-pytorch
A minimal, customizable PyTorch package for building and training convolutional autoencoders based on a simplified U-Net architecture (without skip connections). Ideal for representation learning, image compression, and reconstruction tasks.
## 🔧 Features
- 📦 Modular architecture (`Encoder`, `Decoder`, `AutoEncoder`)
- 🔁 Symmetric U-Net-like design without skip connections
- ⚡ Tanh output activation for stable image reconstruction
- 🧠 Residual blocks with RMS normalization and SiLU activation
- 📱 Designed for image inputs (`3×H×W`) with configurable channels and latent dim
- 🧪 Works with batched input tensors (e.g., `torch.Tensor[B, C, H, W]`)
## 📦 Installation
```bash
pip install convolutional-autoencoder-pytorch
```
## 🧩 Package Structure
```bash
convolutional-autoencoder-pytorch/
├── convolutional_autoencoder_pytorch/
│ ├── __init__.py
│ └── module.py # All architecture classes and logic
├── pyproject.toml
├── LICENSE
└── README.md
```
## 🚀 Quick Start
### 1. Import the package and create the model
```python
import torch
from convolutional_autoencoder_pytorch import AutoEncoder
model = AutoEncoder(
dim=64,
dim_mults=(1, 2, 4, 8),
dim_latent=128,
image_channels=3
)
```
### 2. Forward pass and reconstruction
```python
images = torch.randn(8, 3, 128, 128) # batch of images
reconstructed, latent = model(images)
# Or just get the reconstruction
recon = model.reconstruct(images)
```
### 3. Training step (sample loop)
```python
import torch.nn.functional as F
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
def train_step(images):
model.train()
optimizer.zero_grad()
recon, _ = model(images)
loss = F.mse_loss(recon, images)
loss.backward()
optimizer.step()
return loss.item()
```
## ⚙️ Configuration Options
| Parameter | Description | Default |
|--|--|--|
| `dim` | Base channel size | `64` |
| `dim_mults` | List of multipliers for down/up blocks | `(1, 2, 4, 8)` |
| `dim_latent` | Latent bottleneck dimension | `64` |
| `image_channels` | Input/output image channels (e.g., 3) | `3` |
| `dropout` | Dropout probability | `0.0` |
## 🙋♂️ Author
Developed by [Mehran Bazrafkan](mailto:mhrn.bzrafkn.dev@gmail.com)
This project is an original implementation of a simplified autoencoder architecture. Some ideas and design inspirations were drawn from the open-source [`denoising-diffusion-pytorch`](https://github.com/lucidrains/denoising-diffusion-pytorch) project by Phil Wang, but the code and architecture were written independently.
## 📢 Contributions & Feedback
Contributions, issues, and feedback are welcome via [GitHub Issues](https://github.com/MehranBazrafkan/convolutional-autoencoder-pytorch/issues).
## 📄 License
This project is licensed under the terms of the [MIT LICENSE](https://github.com/MehranBazrafkan/convolutional-autoencoder-pytorch/blob/main/LICENSE).
Raw data
{
"_id": null,
"home_page": null,
"name": "convolutional-autoencoder-pytorch",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.10",
"maintainer_email": null,
"keywords": "autoencoder, pytorch, ml, deep learning",
"author": null,
"author_email": "Mehran Bazrafkan <mhrn.bzrafkn.dev@gmail.com>",
"download_url": "https://files.pythonhosted.org/packages/a8/2f/80804c276228810e07c537f06a8191197a47a42e72583b18642d20de3b53/convolutional_autoencoder_pytorch-0.1.0.tar.gz",
"platform": null,
"description": "\r\n# convolutional-autoencoder-pytorch\r\n\r\nA minimal, customizable PyTorch package for building and training convolutional autoencoders based on a simplified U-Net architecture (without skip connections). Ideal for representation learning, image compression, and reconstruction tasks.\r\n\r\n## \ud83d\udd27 Features\r\n\r\n- \ud83d\udce6 Modular architecture (`Encoder`, `Decoder`, `AutoEncoder`)\r\n- \ud83d\udd01 Symmetric U-Net-like design without skip connections\r\n- \u26a1 Tanh output activation for stable image reconstruction\r\n- \ud83e\udde0 Residual blocks with RMS normalization and SiLU activation\r\n- \ud83d\udcf1 Designed for image inputs (`3\u00d7H\u00d7W`) with configurable channels and latent dim\r\n- \ud83e\uddea Works with batched input tensors (e.g., `torch.Tensor[B, C, H, W]`)\r\n\r\n## \ud83d\udce6 Installation\r\n\r\n```bash\r\npip install convolutional-autoencoder-pytorch\r\n\r\n```\r\n\r\n## \ud83e\udde9 Package Structure\r\n\r\n```bash\r\nconvolutional-autoencoder-pytorch/\r\n\u251c\u2500\u2500 convolutional_autoencoder_pytorch/\r\n\u2502 \u251c\u2500\u2500 __init__.py\r\n\u2502 \u2514\u2500\u2500 module.py # All architecture classes and logic\r\n\u251c\u2500\u2500 pyproject.toml\r\n\u251c\u2500\u2500 LICENSE\r\n\u2514\u2500\u2500 README.md\r\n\r\n```\r\n\r\n## \ud83d\ude80 Quick Start\r\n\r\n### 1. Import the package and create the model\r\n\r\n```python\r\nimport torch\r\nfrom convolutional_autoencoder_pytorch import AutoEncoder\r\n\r\nmodel = AutoEncoder(\r\n dim=64,\r\n dim_mults=(1, 2, 4, 8),\r\n dim_latent=128,\r\n image_channels=3\r\n)\r\n\r\n```\r\n\r\n### 2. Forward pass and reconstruction\r\n\r\n```python\r\nimages = torch.randn(8, 3, 128, 128) # batch of images\r\nreconstructed, latent = model(images)\r\n\r\n# Or just get the reconstruction\r\nrecon = model.reconstruct(images)\r\n\r\n```\r\n\r\n### 3. Training step (sample loop)\r\n\r\n```python\r\nimport torch.nn.functional as F\r\noptimizer = torch.optim.Adam(model.parameters(), lr=1e-4)\r\n\r\ndef train_step(images):\r\n model.train()\r\n optimizer.zero_grad()\r\n recon, _ = model(images)\r\n loss = F.mse_loss(recon, images)\r\n loss.backward()\r\n optimizer.step()\r\n return loss.item()\r\n\r\n```\r\n\r\n## \u2699\ufe0f Configuration Options\r\n\r\n| Parameter | Description | Default |\r\n|--|--|--|\r\n| `dim` | Base channel size | `64` |\r\n| `dim_mults` | List of multipliers for down/up blocks | `(1, 2, 4, 8)` |\r\n| `dim_latent` | Latent bottleneck dimension | `64` |\r\n| `image_channels` | Input/output image channels (e.g., 3) | `3` |\r\n| `dropout` | Dropout probability | `0.0` |\r\n\r\n## \ud83d\ude4b\u200d\u2642\ufe0f Author\r\n\r\nDeveloped by [Mehran Bazrafkan](mailto:mhrn.bzrafkn.dev@gmail.com)\r\nThis project is an original implementation of a simplified autoencoder architecture. Some ideas and design inspirations were drawn from the open-source [`denoising-diffusion-pytorch`](https://github.com/lucidrains/denoising-diffusion-pytorch) project by Phil Wang, but the code and architecture were written independently.\r\n\r\n## \ud83d\udce2 Contributions & Feedback\r\n\r\nContributions, issues, and feedback are welcome via [GitHub Issues](https://github.com/MehranBazrafkan/convolutional-autoencoder-pytorch/issues).\r\n\r\n## \ud83d\udcc4 License\r\n\r\nThis project is licensed under the terms of the [MIT LICENSE](https://github.com/MehranBazrafkan/convolutional-autoencoder-pytorch/blob/main/LICENSE).\r\n",
"bugtrack_url": null,
"license": null,
"summary": "A package to simplify the implementing an autoencoder model.",
"version": "0.1.0",
"project_urls": null,
"split_keywords": [
"autoencoder",
" pytorch",
" ml",
" deep learning"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "330c278f8a4b22b691f32e259b291a52b1d6bf378a8d4fd928d5de2bcb975161",
"md5": "08a2c6e6738b03df5f9e2db3606e374e",
"sha256": "f57d197c9fb366ab48a6454ffd9e96a462aa1cbbd2520fd24574a36e259976f7"
},
"downloads": -1,
"filename": "convolutional_autoencoder_pytorch-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "08a2c6e6738b03df5f9e2db3606e374e",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10",
"size": 5357,
"upload_time": "2025-07-30T11:41:23",
"upload_time_iso_8601": "2025-07-30T11:41:23.747729Z",
"url": "https://files.pythonhosted.org/packages/33/0c/278f8a4b22b691f32e259b291a52b1d6bf378a8d4fd928d5de2bcb975161/convolutional_autoencoder_pytorch-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "a82f80804c276228810e07c537f06a8191197a47a42e72583b18642d20de3b53",
"md5": "54fdc46671e547fc6d2b7ec6a9447574",
"sha256": "360a08e30f43a63079743c55d4ad8dc66b5027a86653738af0160bad0635cc08"
},
"downloads": -1,
"filename": "convolutional_autoencoder_pytorch-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "54fdc46671e547fc6d2b7ec6a9447574",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10",
"size": 4607,
"upload_time": "2025-07-30T11:41:25",
"upload_time_iso_8601": "2025-07-30T11:41:25.350614Z",
"url": "https://files.pythonhosted.org/packages/a8/2f/80804c276228810e07c537f06a8191197a47a42e72583b18642d20de3b53/convolutional_autoencoder_pytorch-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-07-30 11:41:25",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "convolutional-autoencoder-pytorch"
}
JSON