Practical overview
When it fits
Use the repository when you need research code, pretrained weights or a documented ImageNet-1K evaluation and training workflow for EdgeNeXt.
Who it is for
- Computer-vision researchers comparing compact CNN-Transformer architectures.
- Python practitioners evaluating pretrained EdgeNeXt variants on ImageNet-1K.
- Teams exploring an EdgeNeXt model as a starting point for mobile or edge vision work.
Documented uses
- Compare listed model variants by parameter count, multiply-adds and reported ImageNet-1K accuracy.
- Evaluate a pretrained EdgeNeXt checkpoint on ImageNet-1K.
- Prepare the documented environment and dataset for EdgeNeXt-S training.
Requirements & limits
The repository is public and identified as MIT-licensed.
- The model zoo ranges from a 1.33M-parameter, 261M multiply-add variant to an 18.51M-parameter, 3.84G multiply-add variant.
- The documented evaluation workflow requires ImageNet-1K data and pretrained weights.
- The training example specifies a single machine with eight GPUs.
- The repository documents model code and research workflows, not a hosted inference API or mobile runtime.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Evaluate a pretrained EdgeNeXt model on ImageNet-1K
Run the repository’s documented evaluation command with pretrained EdgeNeXt-S weights.
Before you start
- Python 3.8 in a Conda environment
- PyTorch, torchvision and dependencies from requirements.txt
- ImageNet-1K data arranged in train and val class folders
- The edgenext_small pretrained weights
Steps
Create and activate a Conda environment using Python 3.8.
Install PyTorch and torchvision, then install the packages listed in requirements.txt.
Prepare ImageNet-1K with train and val directories, each containing class folders.
Download edgenext_small.pth and run the documented main.py evaluation command with --model edgenext_small, --eval True, --data_path, --output_dir and --resume.
Expected outcome
The repository’s example reports ImageNet-1K top-1 and top-5 accuracy and loss; its listed EdgeNeXt-S result is 79.412 top-1 and 94.512 top-5.
Guide limitations
- The example is specific to ImageNet-1K evaluation and the edgenext_small checkpoint.
- The repository says USI checkpoints require usi_eval True.
- The reported result is the repository’s example, not a guarantee for other data or environments.
Set up the documented multi-GPU training environment
Prepare the repository’s Python environment and ImageNet-1K directory for its documented training workflow.
Before you start
- Conda and Python 3.8
- PyTorch, torchvision and dependencies from requirements.txt
- ImageNet-1K arranged in train and val class folders
- A single machine with eight GPUs for the README’s training command
Steps
Create and activate a Conda environment using Python 3.8.
Install PyTorch and torchvision, then install requirements.txt dependencies.
Prepare the ImageNet-1K directory with train and val folders and class subdirectories.
Expected outcome
The documented environment and dataset are prepared for the repository’s EdgeNeXt-S training command.
Guide limitations
- This guide covers setup only; it does not claim a training run has been executed.
- The README’s training recipe specifies a single machine with eight GPUs.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- mmaaz60/EdgeNeXt · Official homepage title · retained in its source language