Patches Are All You Need?
Paper • 2201.09792 • Published
ConvMixer model
The ConvMixer model is trained on Cifar10 dataset and is based on the paper, github.
Disclaimer : This is a demo model for Sayak Paul's keras example. Please refrain from using this model for any other purpose.
Description
The paper uses 'patches' (square group of pixels) extracted from the image, which has been done in other Vision Transformers like ViT. One notable dawback of such architectures is the quadratic runtime of self-attention layers which takes a lot of time and resources to train for usable output. The ConvMixer model, instead uses Convolutions along with the MLP-mixer to obtain similar results to that of transformers at a fraction of cost.
Intended Use
This model is intended to be used as a demo model for keras-io.
- Downloads last month
- 2
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support