igie

ConvNeXt Small (IGIE)

Model Description

The ConvNeXt Small model represents a significant stride in the evolution of convolutional neural networks (CNNs), introduced by researchers at Facebook AI Research (FAIR) and UC Berkeley. It is part of the ConvNeXt family, which challenges the dominance of Vision Transformers (ViTs) in the realm of visual recognition tasks.

Supported Environments

GPU	IXUCA SDK	Release	Branch
MR-V100	4.4.0	26.03	release/26.03
MR-V100	4.3.0	25.12	release/25.12

Note: 请切换到与您的 SDK 版本对应的 Release 分支进行测试。请勿直接在 master 分支上运行测试，因为 master 分支可能包含与您的本地 SDK 版本不兼容的最新更改。

切换分支命令示例：git checkout release/26.03

Model Preparation

Prepare Resources

Pretrained model: https://download.pytorch.org/models/convnext_small-0c510722.pth

Dataset: https://www.image-net.org/download.php to download the validation dataset.

Install Dependencies

pip3 install -r ../../igie_common/requirements.txt

Model Conversion

python3 ../../igie_common/export.py --model-name convnext_small --weight convnext_small-0c510722.pth --output convnext_small.onnx

Model Inference

export DATASETS_DIR=/Path/to/imagenet_val/
export RUN_DIR=../../igie_common/

FP16

# Accuracy
bash scripts/infer_convnext_small_fp16_accuracy.sh
# Performance
bash scripts/infer_convnext_small_fp16_performance.sh

Model Results

Model	BatchSize	Precision	FPS	Top-1(%)	Top-5(%)
ConvNeXt Small	32	FP16	725.437	83.267	96.515