【第十五届中国计算机设计大赛】智慧导盲组，手把手教你超越0.8

# 克隆PaddleDetection仓库# 如果已经克隆，则不需要重复运行，可把第3行直接注释，从第6行开始运行!git clone https://gitee.com/PaddlePaddle/PaddleDetection.git# 安装其他依赖%cd PaddleDetection
!pip install -r requirements.txt# 编译安装paddledet!python setup.py install
%cd ~

# 查看COCO标注文件import json
coco_anno = json.load(open('WisdomGuide/annotations/instance_train.json'))# coco_anno.keysprint('\nkeys:', coco_anno.keys())# 查看类别信息print('\n物体类别:', coco_anno['categories'])# 查看一共多少张图print('\n图像数量：', len(coco_anno['images']))# 查看一共多少个目标物体print('\n标注物体数量：', len(coco_anno['annotations']))# 查看一条目标物体标注信息print('\n查看一条目标物体标注信息：', coco_anno['annotations'][0])

# 解压数据集!gzip -dfq /home/aistudio/data/data137625/WisdomGuide.tar.gz
!tar -xf /home/aistudio/data/data137625/WisdomGuide.tar -C /home/aistudio/ #把数据集移到/home/aistudio下面

metric: COCOnum_classes: 5TrainDataset:
  !COCODataSet
    image_dir: train
    anno_path: annotations/instance_train.json
    dataset_dir: /home/aistudio/WisdomGuide/
    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']EvalDataset:
  !COCODataSet
    image_dir: val
    anno_path: annotations/instance_val.json
    dataset_dir: /home/aistudio/WisdomGuide/TestDataset:
  !ImageFolder
    ! anno_path: annotations/instance_val.json
    image_dir: val
    dataset_dir: /home/aistudio/WisdomGuide/

# 模型训练%cd ~
%cd PaddleDetection
!python tools/train.py -c configs/picodet/picodet_m_320_coco_lcnet.yml

%cd ~
%cd PaddleDetection# 更换"--infer_img"里的图片路径以预测不同的图片!python tools/infer.py -c configs/picodet/picodet_m_320_coco_lcnet.yml \
                    --infer_img=/home/aistudio/WisdomGuide/val/no_light_629.png \
                    --output_dir=infer_output/ \
                    --draw_threshold=0.5 \
                    -o weights=/home/aistudio/PaddleDetection/output/picodet_m_320_coco_lcnet/best_model.pdparams \
                    --use_vdl=Ture

%cd ~
%cd PaddleDetection# 将"-o weights"里的模型路径换成你自己训好的模型!export CUDA_VISIBLE_DEVICES=0!python tools/export_model.py -c configs/picodet/picodet_m_320_coco_lcnet.yml \
                    -o weights=/home/aistudio/PaddleDetection/output/picodet_m_320_coco_lcnet/best_model.pdparams \
                    TestReader.fuse_normalize=true

| -- model| | -- xxx.pb| | … 
| -- env| | -- 
| |…| -- predict.py| -- …

%cd ~# 在work目录下整理提交代码# 创建model文件夹（已经创建好的话，就不用再创建了）!mkdir /home/aistudio/work/model# 将检测模型拷贝到model文件夹中!cp -r /home/aistudio/PaddleDetection/output_inference/picodet_m_320_coco_lcnet/ /home/aistudio/work/model/# 将训练脚本拷贝到work目录下，与model同级!cp /home/aistudio/PaddleDetection/tools/train.py /home/aistudio/work/# 将PaddleDetection也复制进来!cp -r /home/aistudio/PaddleDetection/ /home/aistudio/work/

# 删除隐藏文件!rm -rf /home/aistudio/work/PaddleDetection/.git/

from .keypoint_preprocess import get_affine_transform

Traceback (most recent call last):
  File "predict.py", line 19, in <module>    from deploy.python.preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride
  File "PaddleDetection/deploy/python/preprocess.py", line 17, in <module>    from keypoint_preprocess import get_affine_transform
ModuleNotFoundError: No module named 'keypoint_preprocess'

# 打包代码%cd /home/aistudio/work/
!zip -r -q -o submission.zip model/ PaddleDetection/ train.py predict.py