- Linux
- Python 3.7
- paddlepaddle-gpu 2.0.2+
- OS: Ubuntu 16.04.6 LTS
- CUDA: 10.2
- CUDNN: 7.6.5
#切换至Dockerfile目录
cd Ultra_light_OCR_No.4#生成镜像
docker build -t paddleocr:gpu .#运行镜像
nvidia-docker run --name Ultra_light_OCR_No.4 -it paddleocr:gpu /bin/bashdata
├── trainval
│ ├── TrainImages
│ │ ├── Train_000000.jpg
│ │ ├── Train_000001.jpg
| | ├── ......
│ ├── LabelTrain.txt
│── testA
│ │ ├── TestAImages
│── testB
│ │ ├── TestBImages
- step1
python3.7 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/new_baseline/step1.yml- step2:
python3.7 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/new_baseline/step2.yml -o Global.checkpoints=checkpoint/step1/iter_epoch_1700# 静态图模型存储位置"./checkpoint/upload_model/infer"
python3.7 tools/infer/predict_rec_media_smart.py \
--rec_config=./configs/rec/new_baseline/step2.yml \
--rec_model_dir=./checkpoint/upload_model/infer \
--image_dir=./data/testB/TestBImages \
--save_res_path=./output/rec/testB.txt