TopFormer: Token Pyramid Transformer for Mobile Semantic Segmentation, arxiv

PaddlePaddle training/validation code and pretrained models for the model released in CVPR2022: TopFormer (classification backbone).

The official PyTorch implementation is here.

This implementation is developed by PaddleViT.

TopFormer Model Overview

Update

Update (2022-06-28): Model weight trained from scracth using paddlevit is uploaded.
Update (2022-04-22): Code is released and ported weights are uploaded.

Models Zoo

Model	Acc@1	Acc@5	#Params	FLOPs	Image Size	Crop_pct	Interpolation	Link
topformer_tiny	65.98	87.32	1.5M	0.13G	224	0.875	bicubic	google/baidu
topformer_small	72.44	91.17	3.1M	0.24G	224	0.875	bicubic	google/baidu
topformer_base	75.25	92.67	5.1M	0.37G	224	0.875	bicubic	google/baidu

*The results are above are ported from official implemetation and evaluated on ImageNet2012 validation set.

Model weights trained from scratch using PaddleViT are also provided:

Model	Acc@1	Acc@5	#Params	FLOPs	Image Size	Crop_pct	Interpolation	Link	Log
topformer_tiny	67.63	87.82	1.5M	0.13G	224	0.875	bicubic	google/baidu	google/baidu
topformer_small	72.61	90.78	3.1M	0.24G	224	0.875	bicubic	google/baidu	google/baidu
topformer_base	74.22	91.74	5.1M	0.37G	224	0.875	bicubic	google/baidu	google/baidu

Note: accuracy is validated using model EMA.

Data Preparation

ImageNet2012 dataset is used in the following file structure:

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

train_list.txt: list of relative paths and labels of training images. You can download it from: google/baidu
val_list.txt: list of relative paths and labels of validation images. You can download it from: google/baidu

Usage

To use the model with pretrained weights, download the .pdparam weight file and change related file paths in the following python scripts. The model config files are located in ./configs/.

For example, assume weight file is downloaded in ./topformer_tiny.pdparams, to use the topformer_tiny model in python:

from config import get_config
from topformer import build_topformer as build_model
# config files in ./configs/
config = get_config('./configs/topformer_tiny.yaml')
# build model
model = build_model(config)
# load pretrained weights
model_state_dict = paddle.load('./topformer_tiny.pdparams')
model.set_state_dict(model_state_dict)

Evaluation

To evaluate model performance on ImageNet2012, run the following script using command line:

sh run_eval_multi.sh

or

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
python main_multi_gpu.py \
-cfg='./configs/topformer_tiny.yaml' \
-dataset='imagenet2012' \
-batch_size=256 \
-data_path='/dataset/imagenet' \
-eval \
-pretrained='./topformer_tiny.pdparams' \
-amp

Note: if you have only 1 GPU, change device number to CUDA_VISIBLE_DEVICES=0 would run the evaluation on single GPU.

Training

To train the model on ImageNet2012, run the following script using command line:

sh run_train_multi.sh

or

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
python main_multi_gpu.py \
-cfg='./configs/topformer_tiny.yaml' \
-dataset='imagenet2012' \
-batch_size=256 \
-data_path='/dataset/imagenet' \
-amp

Note: it is highly recommanded to run the training using multiple GPUs / multi-node GPUs.

Reference

@article{zhang2022topformer,
  title={TopFormer: Token Pyramid Transformer for Mobile Semantic Segmentation},
  author={Zhang, Wenqiang and Huang, Zilong and Luo, Guozhong and Chen, Tao and Wang, Xinggang and Liu, Wenyu and Yu, Gang and Shen, Chunhua},
  journal={arXiv preprint arXiv:2204.05525},
  year={2022}
}