Skip to content

Latest commit

 

History

History
72 lines (55 loc) · 2.94 KB

README.md

File metadata and controls

72 lines (55 loc) · 2.94 KB
Raw

CaiT

PaddlePaddle reimplementation of facebookresearch's repository for the cait model that was released with the paper CaiT: Going deeper with Image Transformers.

Requirements

To enjoy some new features, PaddlePaddle 2.4 is required. For more installation tutorials refer to installation.md

How to Train

# Note: Set the following environment variables 
# and then need to run the script on each node.
export PADDLE_NNODES=1
export PADDLE_MASTER="xxx.xxx.xxx.xxx:12538"
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7

python -m paddle.distributed.launch \
    --nnodes=$PADDLE_NNODES \
    --master=$PADDLE_MASTER \
    --devices=$CUDA_VISIBLE_DEVICES \
    plsc-train \
    -c ./configs/cait_s24_224_in1k_1n8c_dp_fp16o2.yaml

How to Evaluation

# [Optional] Download checkpoint
mkdir -p pretrained/
wget -O ./pretrained/cait_s24_224_in1k_1n8c_dp_fp16o2.pdparams https://plsc.bj.bcebos.com/models/cait/v2.4/cait_s24_224_in1k_1n8c_dp_fp16o2.pdparams

export PADDLE_NNODES=1
export PADDLE_MASTER="127.0.0.1:12538"
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python -m paddle.distributed.launch \
  --nnodes=$PADDLE_NNODES \
  --master=$PADDLE_MASTER \
  --devices=$CUDA_VISIBLE_DEVICES \
  plsc-eval \
  -c ./configs/cait_s24_224_in1k_1n8c_dp_fp16o2.yaml \
  -o Global.pretrained_model=pretrained/cait_s24_224_in1k_1n8c_dp_fp16o2 \
  -o Global.finetune=False

Other Configurations

We provide more directly runnable configurations, see CaiT Configurations.

Models

Model Phase Dataset Configs GPUs Img/sec Top1 Acc Pre-trained checkpoint Fine-tuned checkpoint Log
cait_s24_224 pretrain ImageNet2012 config A100*N1C8 2473 0.82628 download log

Citations

@InProceedings{Touvron_2021_ICCV,
    author    = {Touvron, Hugo and Cord, Matthieu and Sablayrolles, Alexandre and Synnaeve, Gabriel and J\'egou, Herv\'e},
    title     = {Going Deeper With Image Transformers},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {32-42}
}