edgenext

EdgeNeXt

EdgeNeXt: Efficiently Amalgamated CNN-Transformer Architecture for Mobile Vision Applications

Introduction

EdgeNeXt effectively combines the strengths of both CNN and Transformer models and is a new efficient hybrid architecture. EdgeNeXt introduces a split depth-wise transpose attention (SDTA) encoder that splits input tensors into multiple channel groups and utilizes depth-wise convolution along with self-attention across channel dimensions to implicitly increase the receptive field and encode multi-scale features.[1]

Figure 1. Architecture of EdgeNeXt [1]

Requirements

mindspore	ascend driver	firmware	cann toolkit/kernel
2.3.1	24.1.RC2	7.3.0.1.231	8.0.RC2.beta1

Quick Start

Preparation

Installation

Please refer to the installation instruction in MindCV.

Dataset Preparation

Please download the ImageNet-1K dataset for model training and validation.

Training

• Distributed Training

It is easy to reproduce the reported results with the pre-defined training recipe. For distributed training on multiple Ascend 910 devices, please run

# distributed training on multiple NPU devices
msrun --bind_core=True --worker_num 8 python train.py --config configs/edgenext/edgenext_small_ascend.yaml --data_dir /path/to/imagenet

For detailed illustration of all hyper-parameters, please refer to config.py.

Note: As the global batch size (batch_size x num_devices) is an important hyper-parameter, it is recommended to keep the global batch size unchanged for reproduction or adjust the learning rate linearly to a new global batch size.

• Standalone Training

If you want to train or finetune the model on a smaller dataset without distributed training, please run:

# standalone training on single NPU device
python train.py --config configs/edgenext/edgenext_small_ascend.yaml --data_dir /path/to/dataset --distribute False

Validation

To validate the accuracy of the trained model, you can use validate.py and parse the checkpoint path with --ckpt_path.

python validate.py -c configs/edgenext/edgenext_small_ascend.yaml --data_dir /path/to/imagenet --ckpt_path /path/to/ckpt

Performance

Our reproduced model performance on ImageNet-1K is reported as follows.

Experiments are tested on ascend 910* with mindspore 2.3.1 graph mode.

model name	params(M)	cards	batch size	resolution	jit level	graph compile	ms/step	img/s	acc@top1	acc@top5	recipe	weight
edgenext_xx_small	1.33	8	256	256x256	O2	389s	239.38	8555.43	70.64	89.75	yaml	weights

Experiments are tested on ascend 910 with mindspore 2.3.1 graph mode.

model name	params(M)	cards	batch size	resolution	jit level	graph compile	ms/step	img/s	acc@top1	acc@top5	recipe	weight
edgenext_xx_small	1.33	8	256	256x256	O2	311s	191.24	10709.06	71.02	89.99	yaml	weights

Notes

• top-1 and top-5: Accuracy reported on the validation set of ImageNet-1K.

References

[1] Maaz M, Shaker A, Cholakkal H, et al. EdgeNeXt: efficiently amalgamated CNN-transformer architecture for Mobile vision applications[J]. arXiv preprint arXiv:2206.10589, 2022.