[Reproduced] A detailed understanding of the use of TensorFlow object detection API! ! (Help in-depth understanding)

1. TensorFlow training CNN classifier
2. TensorFlow-slim training CNN classification model
3. TensorFlow custom generates .record files
4. TensorFlow-slim training CNN classification model (continued)
5. TensorFlow custom model export: convert .ckpt format to .pb format
6. TensorFlow model saving and restoration

• One advantage of saving as a .ckpt model is that when you need to continue training, you only need to import the trained model results, and then continue training on this basis. The following .pb format cannot continue training, because the model parameters saved in this format have been converted to constants (not variables).

7. TensorFlow uses the pre-trained model ResNet-50 (continued)

• Use the pre-training model to fine-tune the neural network needs to do: 1. Define the neural network structure; 2. Import the pre-training model parameters; 3. Read the data for training; 4. Use Tensorboard to visualize the training process (omitted here, left to I will talk about it separately later).