Skip to content

Latest commit

 

History

History
53 lines (35 loc) · 1.44 KB

README.md

File metadata and controls

53 lines (35 loc) · 1.44 KB

Tensorflow_detection_tutorial

在/usr/local/models工程下新建文件夹dataset和data:

image

image

其中xml_to_csv.py和generate_tfrecord.py可在master中寻找

注意修改

def class_text_to_int(row_label): if row_label == 'car': return 1 elif row_label == 'person': return 2 elif row_label == 'truck': return 3 elif row_label == 'bus': return 4 elif row_label == 'other vehicle': return 5 else: print('NONE: ' + row_label) # None

以及相关路径

运行完成,在data目录下产生

eval.csv, eval.record, train.csv, train.record

在data下放置labelmap.pbtxt,注意和上述文件内容保持一致

https://github.com/tensorflow/models/tree/master/research/slim 下载模型,将 model.ckpt.data-00000-of-00001 model.ckpt.index model.ckpt.meta 放置在dataset文件夹下的fine_tune_model文件夹内

将下载的pipeline.config放置在dataset/data/文件夹内

修改pipeline.config中的尺寸,fine_tune_checkpoint,num_steps,label_map_path,input_path,num_examples

修改models-master/research/object_detection/model_main.py

注意在dataset下新建training文件夹,保存训练模型。

注意export PYTHONPATH,否则会报错。