InceptionMLP: A Vision MLP Architecture with Multi-branch Features Fusion (in process)

Acc. of Recent MLP-like Models on Imagenet-1K

Model	Sub-model	Parameters	Top 1 Acc.
Cycle-MLP	T	28M	81.3
ViP	Small/7	25M	81.5
iMLP	S	20M	82.1
Hire-MLP	S	33M	82.1
DynaMixer	S	26M	82.7
Cycle-MLP	S	50M	82.9
ViP	Medium/7	55M	82.7
Hire-MLP	B	58M	83.2
DynaMixer	M	57M	83.7
Cycle-MLP	B	88M	83.4
ViP	Large/7	88M	83.2
Hire-MLP	L	96M	83.8

Pre-trained iMLP now available at: Baidu Disk

Requirements

Environment

wandb
torch==1.9.0
torchvision>=0.10.0
pyyaml
timm==0.4.12

Data

ImageNet with the following folder structure

│imagenet/
├──train/
│  ├── n01440764
│  │   ├── n01440764_10026.JPEG
│  │   ├── n01440764_10027.JPEG
│  │   ├── ......
│  ├── ......
├──val/
│  ├── n01440764
│  │   ├── ILSVRC2012_val_00000293.JPEG
│  │   ├── ILSVRC2012_val_00002138.JPEG
│  │   ├── ......
│  ├── ......

Training example

Command line for training on 8 GPUs, parameters can be modified with yaml file

./distributed_train.sh 8 --amp --sync-bn --pin-mem

Thanks

The code is based on former work Dynamixer and AS-MLP