Overview
ConvNeXt is a pure ConvNet model developed by Facebook AI Research and UC Berkeley. It's designed to be accurate, efficient, scalable, and simple for image classification tasks. The architecture focuses on standard ConvNet modules, making it easy to implement and integrate into existing workflows. ConvNeXt models are pre-trained on ImageNet-1K and ImageNet-22K datasets. Fine-tuning code and downstream transfer learning code are available, supporting object detection and semantic segmentation. The repository provides pre-trained models, training code, and evaluation scripts. It leverages PyTorch, timm library, DeiT, and BEiT repositories for implementation.