- https://www.bilibili.com/video/av16001891?from=search&seid=16798140708617054978
- 参考书目 <<TensorFlow 技术解析和实战>>
https://www.cnblogs.com/xianhan/p/9146485.html
https://blog.csdn.net/rookie_wei/article/details/81275663
https://blog.csdn.net/jesmine_gu/article/details/81155787
https://blog.csdn.net/dy_guox/article/details/79111949
- python xmltocsv.py
- Create train data: python generatetfrecords.py --csv_input=data/dataset_train.csv --output_path=train.record python xmltorecod.py --csv_input=data/dataset_train.csv --output_path=train.record python xmltorecod.py --csv_input=data/test_2.csv --output_path=test.record python xmltorecod.py --csv_input=data/train_4.csv --output_path=train.record -(!!!一定要将main里面的路径改为自己的图片们)对应文件夹的images下
- change config ssd_mobilenet_v1_coco.config
- start train
python ./legacy/train.py --logtostderr --train_dir=training2/ --pipeline_config_path=training2/ssd_mobilenet_v1_coco.config --num_train_steps=20000 --num_eval_steps=2000
python export_inference_graph.py \ --input_type image_tensor \ --pipeline_config_path training2/ssd_mobilenet_v1_coco.config \ --trained_checkpoint_prefix training2/model.ckpt-30000 \ --output_directory hand_graph