SKNet¶

SKNet: Selective Kernel Networks

Introduction¶

The core idea of SKNet: SK Convolution

Split is to perform different receptive field convolution operations. The upper branch is a 3 x 3 kernel with dilate size=1, and the lower one is a 3 x 3 convolution with dilate size=2.
Fuse performs feature fusion, superimposes the convolutional features of the two branches, and then performs the standard SE process.
Select is to aggregate the feature maps of differently sized kernels according to the selection weights.

				Pynative	Pynative	Graph	Graph
	Model	Top-1 (%)	Top-5 (%)	train (s/epoch)	Infer (ms)	train(s/epoch)	Infer (ms)	Download	Config
GPU	sknet50							model	config
Ascend	sknet50

The yaml config files that yield competitive results on ImageNet for different models are listed in the configs folder. To trigger training using preset yaml config.
```
comming soon
```

Here is the example for finetuning a pretrained SKNet on CIFAR10 dataset using Adam optimizer.

python train.py --model=sknet50 --pretrained --opt=momentum --lr=0.001 dataset=cifar10 --num_classes=10 --dataset_download

Detailed adjustable parameters and their default value can be seen in config.py.

To validate the model, you can use validate.py. Here is an example to verify the accuracy of pretrained weights.
```
python validate.py --model=sknet50 --dataset=imagenet --val_split=val --pretrained
```

To validate the model, you can use validate.py. Here is an example to verify the accuracy of your training.

python validate.py --model=sknet50 --dataset=imagenet --val_split=val --ckpt_path='./ckpt/sknet50-best.ckpt'