ConvMixer is a simple yet effective vision architecture that combines large-kernel depthwise convolutions for spatial mixing with pointwise convolutions for channel mixing, achieving transformer-like performance with minimal complexity.

Original paper: Patches Are All You Need? ConvMixer

ConvMixer-768/32

This model uses the ConvMixer-768/32 variant, which processes 32×32 patches with 768 feature channels, providing strong accuracy while remaining computationally efficient. It is well suited for image classification tasks where simplicity, speed, and high accuracy are desired, and can serve as a lightweight backbone for research or prototyping.

Model Configuration:

• Reference implementation: Official ConvMixer source code
• Original Weight: Convmixer_768_32_ks7_p7_relu
• Resolution: 3x224x224
• Support Cooper version:
  • Cooper SDK: [2.5.2]
  • Cooper Foundry: [2.2]

Model	Device	Model Link
ConvMixer-768/32	N1-655	Model_Link
ConvMixer-768/32	CV72	Model_Link
ConvMixer-768/32	CV75	Model_Link