English | 简体中文

VIMER-UFO 2.0 (CV Foundation Model)

• Instructions
  • Environment
  • Distributed-training
  • Download the 17B pretrained model
  • Extract task specific model from 17B pretrained model
  • Evaluation

Instructions

Demo is based on 48 A100 cards.

Environment

Please use python3.7 and cuda11.0

pip install -U pip==22.0.3
pip install -r requirements.txt.train
pip install faiss-gpu==1.7.2 --no-cache
pip install paddlepaddle-gpu==0.0.0.post110 -f https://www.paddlepaddle.org.cn/whl/linux/gpu/develop.html

Configure environment variables.
```bash
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export PYTHONPATH=/path/to/AllInOne
export FASTREID_DATASETS=/path/to/data/train_datasets
export UFO_config=configs/MS1MV3_PersonAll_VeriAll_SOP_Decathlon_Intern/vithuge_lr2e-1_iter6w_dpr0.2_moeplusbalance_tasklr_dataaug.py

Distributed-training

mpirun -npernode 1 --bind-to none python -m paddle.distributed.launch  --gpus "0,1,2,3,4,5,6,7" tools/ufo_trainsuper_moe.py  --config-file $UFO_config

Download-17B-pretrained-model

Please send your request to [email protected] . The request may include your name and orgnization. We will notify you by email as soon as possible.

Extract-task-specific-model-from-17B-pretrained-model

Take ImageNet as an example

python tools/extract_task_specific_model.py --paddel_model_path UFO_2.0_17B_release.pdmodel --task_name task6

Evaluation

Take ImageNet as an example

export UFO_config=configs/evaluation_configs/vithuge_imagenet_eval.py
python -m paddle.distributed.launch  --gpus "0,1,2,3,4,5,6,7" tools/ufo_train.py --config-file $UFO_config --eval-only

We thank for https://github.com/facebookresearch/detectron2 and https://github.com/JDAI-CV/fast-reid .

If you have any question, please contact [email protected]