TinyVision is an evolving research project focused on designing ultra-lightweight image classification models with minimal parameter counts. The goal is to explore what’s actually necessary for fundamental vision tasks by combining handcrafted feature preprocessing with highly efficient CNN architectures.
📦 Current Release: v2.0.0
🔖 Zenodo DOI: 10.5281/zenodo.16467349
📁 Latest Results & Code: See the cat_vs_dog_classifier/final/v2 directory
⚠️ This release does not include a paper, but focuses on the codebase, experiment results, and reproducible training scripts. A deeper analysis and formal documentation will come in future updates.
- ✅ Cat vs Dog Classification
First completed task using a 25,000-image dataset with handcrafted preprocessing + compact CNNs.- Achieved up to 86.87% test accuracy with models under 12.5k parameters
- Several models under 5k parameters reached over 83% accuracy, showcasing strong efficiency-performance trade-offs.
- 📂 Final results and code for this task are in the
cat_vs_dog_classifier/final/v2directory.
- 📊 Add thorough performance analysis of model architectures to understand why something works while others don't
- 🧩 Explore new vision tasks (edge detection, object detection, etc.) with compact models
- 📖 Expand documentation, architecture diagrams, and visualizations
- 🧠 Log and reflect on failed or inconclusive experiments critical for understanding design boundaries
This project is currently personal and tracks my ongoing experiments.
I’m not accepting pull requests, but you're welcome to:
- 📬 Open an issue for discussion or feedback
- 💌 Reach me at:
[email protected] - 📢 Follow me on X
Small models aren't just about speed—they’re a design challenge.
How much can we cut before it breaks? What’s essential? What’s fluff?
TinyVision is my attempt to find those answers.