DCGAN Magic on MNIST: Enchanting Image Dataset! 🌟🎨

Dive into a realm where pixels awaken with magic! 🚀✨ This captivating dataset unveils the enchanting journey of a Deep Convolutional Generative Adversarial Network (DCGAN) as it breathes life into mesmerizing images on the MNIST canvas.

🌌 GIF Chronicles: Behold time capsules of GIFs, each a magical journey through epochs. Watch as the pixels evolve, dance, and transform, revealing the growth and artistry of our wizardly model.

📸 Snapshot Diaries: Explore meticulously collected snapshots, capturing the essence of every 1000 steps across 10 enchanting epochs. Each image tells a tale of the model's evolution, from its tentative steps to the grandeur of mastery.

🧙‍♂️ Genesis Moments: Step back to the humble beginnings, where the random generator forged base generations, setting the stage for the grand symphony of creativity.

🎩 Crowning Achievements: Marvel at the final generator's crowning glory—the synthesis of captivating, realistic images. Each pixel is a stroke of genius, a testament to the magic of PyTorch and the artistry of our DCGAN.

May your coding be enchanting, and your images spellbinding! 🪄🌈

May the Pixels Be Ever in Your Favor! 🖼💫

Dataset

This dataset was created by Andrew Parr

Contents

A collection of PyTorch checkpoints for generating abstract art using a DCGAN.

The images are of the dimension: 64x64x3

You can load the checkpoint using:

checkpoint = torch.load(checkpoint_path, map_location=device)

See notebook for architecture implementation: https://www.kaggle.com/code/therealcyberlord/abstract-art-generation-dcgan

Dataset

This dataset was created by Andrew Parr

Contents

Dataset

This dataset was created by Andrew Parr

Contents

Dataset

This dataset was created by Andrew Parr

Contents

Dataset

This dataset was created by Andrew Parr

Contents

Dataset

This dataset was created by Andrew Parr

Contents

Dataset

This dataset was created by Andrew Parr

Contents

📌 Dataset Description

This dataset contains synthetic brain tumor MRI scans generated using a Conditional Deep Convolutional Generative Adversarial Network (CDCGAN). The images were generated to closely resemble real MRI scans from the original Brain Tumor MRI Dataset (sourced from Kaggle).

The dataset is organized into training and testing directories, each with four classes of brain MRI images:

🧠 Classes

Glioma – Synthetic MRI scans of glioma brain tumors

Meningioma – Synthetic MRI scans of meningioma brain tumors

Notumor – Synthetic MRI scans with no brain tumor present

Pituitary – Synthetic MRI scans of pituitary brain tumors

📂 Dataset Structure

Training Set

glioma/ → 1321 synthetic images

meningioma/ → 1339 synthetic images

notumor/ → 1595 synthetic images

pituitary/ → 1457 synthetic images

Testing Set

glioma/ → 300 synthetic images

meningioma/ → 306 synthetic images

notumor/ → 405 synthetic images

pituitary/ → 300 synthetic images

⚙️ Generation Details

Image size: 256 × 256 (grayscale)

Model: Conditional DCGAN (PyTorch)

Training epochs: 64

Synthetic dataset root: /kaggle/working/synthetic_brain_tumor_mri

Metadata file: _synthetic_summary.json (contains generation details, class mappings, and counts)

🎯 Applications

Data augmentation for training deep learning models in medical imaging

Research on GAN-based synthetic data generation

Benchmarking explainability and robustness of AI models

Privacy-preserving medical image synthesis

📜 License

This dataset is released under CC0: Public Domain. It may be freely used, modified, and distributed for research and educational purposes.

GO RICK YOURSELF 👀

This dataset was built using pyautogui on the Adult Swim Rick and Morty character creator website GoRickYourself 💚

I was doing a project to analyze the effectiveness of GAN's trained on cartoon images, and there weren't any datasets out there that fitted my need so I made my own. I purposefully decided to miss some of the really weird features available on the website (spider and monster legs), because I figured it would confuse the model. I also used only the plain backgrounds since the colorful ones would've added extra and unnecessary information. I would be more than happy to create a new dataset without these constraints if anyone would be interested ✔️

The images were used to train StyleGAN2 and a few variants of DCGAN. The StyleGAN model achieved a pretty good FID score of 28 - I was impressed! I hope you enjoy looking at the weird and wonderful characters that I generated, and I would love to see what you make 😎

Disclaimer: I have contacted AdultSwim to ask for permission to publish this dataset but have not yet received a reply. So it will either be here until I get told to take it down 💔 or until it goes viral enough for Dan Harmon and Justin Roiland to see it 😈

And while you're at it, feel free to check out my website ✌️

[I will try add a notebook at some stage]

📌 Dataset Description

This dataset contains synthetic brain tumor MRI scans generated using a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP).

The images were created to resemble real MRI scans from the original Brain Tumor MRI Dataset (sourced from Kaggle) and serve as high-quality synthetic data for medical imaging research.

🧠 Classes

Glioma – Synthetic MRI scans of glioma brain tumors

Meningioma – Synthetic MRI scans of meningioma brain tumors

Notumor – Synthetic MRI scans with no brain tumor present

Pituitary – Synthetic MRI scans of pituitary brain tumors

📂 Dataset Structure Training Set

glioma/ → 1321 synthetic images

meningioma/ → 1339 synthetic images

notumor/ → 1595 synthetic images

pituitary/ → 1457 synthetic images

Testing Set

glioma/ → 300 synthetic images

meningioma/ → 306 synthetic images

notumor/ → 405 synthetic images

pituitary/ → 300 synthetic images

⚙️ Generation Details

Image size: 256 × 256 (grayscale)

Model: WGAN-GP (PyTorch)

Training epochs: 32

Lambda GP: 10

Critic updates (n_critic): 5

Learning rate (generator): 0.0001

Synthetic dataset root: /kaggle/working/synthetic_brain_tumor_mri_wgan_gp

Metadata file: _synthetic_summary.json (contains generation details, class mappings, and counts)

🎯 Applications

Data augmentation for training deep learning models in medical imaging

Research on GAN-based synthetic data generation (WGAN vs. DCGAN/CDCGAN comparison)

Benchmarking explainability and robustness of AI models

Privacy-preserving medical image synthesis for clinical research

📜 License

This dataset is released under CC0: Public Domain. It may be freely used, modified, and distributed for research and educational purposes.