VisionX

PyTorch based repository for structured and modularised code for various Computer Vision models, for results/metrics, ease of use and out of the box experimentation : work in progress

---

How to interpret the repository:

1. Models dir has all CV models, custom models are under custom.py.

2. Utils dir contains utility functions and components.

3. Extra resources like images or gifs would go under resources .

4. Notebooks dir will hold all Jupyter Notebooks for experiments and/or final runs for presenting results.

---

Interesting results/experiments:

• 89K and 95K parameter models on Cifar10 with 85%+ validation accuracy - uses augmentation + groups param
• 93% validation accuracy on Cifar10 with Custom ResNet model - uses augmentation + One Cycle LR