README.md

Semantic Segmentation

Segmentation on ADE20K is implemented based on MMSegmentation. We follow the hyper-parameters of PVT and PoolFormer for the comparison.

Requirements

Install mmcv-full and MMSegmentation v0.19.0. Later versions should work as well. The easiest way is to install via MIM

pip install -U openmim
mim install mmcv-full
mim install mmseg

Data preparation

We benchmark EfficientFormer on the challenging ADE20K dataset, which can be downloaded and prepared following insructions in MMSeg. The data should appear as:

├── segmentation
│   ├── data
│   │   ├── ade
│   │   │   ├── ADEChallengeData2016
│   │   │   │   ├── annotations
│   │   │   │   │   ├── training
│   │   │   │   │   ├── validation
│   │   │   │   ├── images
│   │   │   │   │   ├── training
│   │   │   │   │   ├── validation

Testing

Weights trained on ADE20K can be downloaded here. We provide a multi-GPU testing script, specify config file, checkpoint, and number of GPUs to use:

sh ./tools/dist_test.sh config_file path/to/checkpoint #GPUs --eval mIoU

For example, to test EfficientFormer-L1 on ADE20K on an 8-GPU machine,

sh ./tools/dist_test.sh configs/sem_fpn/fpn_efficientformer_l1_ade20k_40k.py path/to/efficientformer_l1_ade20k.pth 8 --eval mIoU

Training

ImageNet Pretraining

Put ImageNet-1K pretrained weights of backbone as

EfficientFormer
├── weights
│   ├── efficientformer_l1_300d.pth
│   ├── ...

Single machine multi-GPU training

We provide PyTorch distributed data parallel (DDP) training script dist_train.sh, for example, to train EfficientFormer-L1 on an 8-GPU machine:

sh ./tools/dist_train.sh configs/sem_fpn/fpn_efficientformer_l1_ade20k_40k.py 8

Tips: specify configs and #GPUs!

Multi-node training

On Slurm-managed cluster, multi-node training can be launched by slurm_train.sh, similarly, to train EfficientFormer:

sh ./tools/slurm_train.sh your-partition exp-name config-file 

Tips: specify GPUs/CPUs/memory per node in the script slurm_train.sh based on your resource!