hello everyone!
i have been building Refrakt for the past few months, a workflow for training and evaluating computer vision models.
deep learning models today are fragmented:
- training usually lives in one place.
- evaluation lives somewhere else,
- and explainability is usually considered last.
Refrakt is a unified platform that brings all of these elements into a single system.
i've put together a walkthrough video where you can understand more about it: Refrakt: A Unified Platform for Deep Learning Workflows
if you would like to wait for the full platform access: Refrakt
if you would like to run your own configuration for training, follow this format in the demo:
model: resnet18 (more models coming soon)
dataset:
source: torchvision (only torchvision models supported right now)
name: CIFAR10 (or MNIST)
mode: train
device: auto
setup: quick (for 2 epochs, or 5 for full training)
i would love your thoughts and gather your feedback so that Refrakt can be a better product for people to use.
