参考操作:
https://blog.csdn.net/wuhe671767/article/details/82697303
问题解决:
成功解决object_detection/protos/*.proto: No such file or directory
https://blog.csdn.net/qq_41185868/article/details/82889473
win10下用tensorflow object detection API 验证时报No module named 'object_detection'
https://blog.csdn.net/whitesilence/article/details/79047107
Windows 10 编译 Pycocotools 踩坑记
https://www.jianshu.com/p/de455d653301
Tensorflow object detection API QA合集
https://blog.csdn.net/weixin_41997327/article/details/85232056
使用tensorboard 查看图时报 ValueError: Invalid format string
https://blog.csdn.net/michealblank/article/details/88085243
AttributeError: '_NamespacePath' object has no attribute 'sort'
https://blog.csdn.net/appleaoliao/article/details/82796469
InvalidArgumentError
Assign requires shapes of both tensors to match.
删除旧的checkpoint,这是因为我把tensorflow的gpu版本卸载了的原因。
can't import name 'abs'
第一步:位于我的训练文件夹中
# 数据准备:
+ images文件夹
+ annotations文件夹{labelImage.py生成}
+ train.txt(+val.txt){由list.py生成或者make_main_txt.py一键生成}
+ raccoon_label_map.pbtxt {改编号}
# 数据制作格式(要train.txt和val.txt各制作一次 + 标签类别制作--label_map_path):
python create_pascal_tf_record4raccoon.py --data_dir=./images --set=./train.txt --output_path=./train.record --label_map_path=./raccoon_label_map.pbtxt --annotations_dir=./annotations
# 如果输出 if not xml,说明执行成功
#具体实现:
python create_pascal_tf_record4raccoon.py --data_dir=./image --set=./train.txt --output_path=./train.record --label_map_path=./raccoon_label_map.pbtxt --annotations_dir=./annotations
python create_pascal_tf_record4raccoon.py --data_dir=./image --set=./val.txt --output_path=./val.record --label_map_path=./raccoon_label_map.pbtxt --annotations_dir=./annotations
第二步:【训练】 位于API文件夹中
# 打开终端 + 训练
+(ssd_mobilenet_v1_raccoon.config
+ 编辑微调模型、
+ 输入类别数、
+ 训练和验证样本的地址)
python train.py --logtostderr --pipeline_config_path=./raccoon_dataset-master/ssd_mobilenet_v1_raccoon.config --train_dir=./train
具体实现:
python train.py --logtostderr --pipeline_config_path=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset-master/ssd_mobilenet_v1_raccoon.config --train_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset-master/train
python train.py --logtostderr --pipeline_config_path=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/ssd_mobilenet_v1_raccoon.config --train_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/train
# 位于我的训练文件夹中
# 【查看训练过程】训练结果的文件夹
tensorflow.exe --logdir=./
第三步:【验证】位于API文件夹中
python eval.py --logtostderr --pipeline_config_path=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset-master/ssd_mobilenet_v1_raccoon.config --checkpoint_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset-master/train --eval_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset-master/eval
python eval.py --logtostderr --pipeline_config_path=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/ssd_mobilenet_v1_raccoon.config --checkpoint_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/train --eval_dir=C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/eval
第四步:【模型转换】位于API文件夹中-最终保存至我的训练文件夹的model_translate文件夹
python export_inference_graph.py --input_type image_tensor --pipeline_config_path ./ssd_mobilenet_v1_raccoon.config --trained_checkpoint_prefix ./train/model.ckpt-39527 --output_directory ./model_translate/
python export_inference_graph.py --input_type image_tensor --pipeline_config_path C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/ssd_mobilenet_v1_raccoon.config --trained_checkpoint_prefix C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/train/model.ckpt-12221 --output_directory C:/Users/Administrator/Desktop/demo/DL/demo_6/raccoon_dataset_3c/model_translate/
第五步:位于我的训练文件夹中
# 模型调用使用
在test文件夹中
生成的模型model_translate + raccoon_label_map.pbtxt标签名 + 最终结果保存文件夹result
+ 编辑检测脚本test.py的 类别数量NUM_CLASSES = n类