README.md

---
library_name: pytorch
---

![convnext_logo](resource/ConvNeXt.png)

ConvNeXt revisits and modernizes convolutional neural network design by incorporating architectural insights from Vision Transformers, such as large kernels, simplified blocks, and improved normalization, while retaining convolutional efficiency.

Original paper: [A ConvNet for the 2020s, Liu et al., 2022](https://arxiv.org/abs/2201.03545)

# ConvNeXt-T

This model uses the ConvNeXt-Tiny variant, a lightweight configuration that delivers strong accuracy with relatively low computational cost. It is well suited for high-resolution image classification and as a general-purpose backbone for detection and segmentation tasks where CNN efficiency is preferred.

Model Configuration:
- Reference implementation: [ConvNeXt_T](https://pytorch.org/vision/stable/models/generated/torchvision.models.convnext_tiny.html)
- Original Weight: [ConvNeXt_Tiny_Weights.IMAGENET1K_V1](https://download.pytorch.org/models/convnext_tiny-983f1562.pth)
- Resolution: 3x224x224
- Support Cooper version:
    - Cooper SDK: [2.5.2]
    - Cooper Foundry: [2.2]

| Model | Device | Model Link |
| :-----: | :-----: | :-----: |
| ConvNeXt-T | N1-655 | [Model_Link](https://huggingface.co/Ambarella/ConvNeXt/blob/main/n1-655_convnext_tiny.bin) |
| ConvNeXt-T | CV72 | [Model_Link](https://huggingface.co/Ambarella/ConvNeXt/blob/main/cv72_convnext_tiny.bin) |
| ConvNeXt-T | CV75 | [Model_Link](https://huggingface.co/Ambarella/ConvNeXt/blob/main/cv75_convnext_tiny.bin) |