TensorFlow-Slim 图片分类（二）

TensorFlow-Slim图片分类 TensorFlow-Slim image classification

1 将图片数据转换成TF-Record格式文件 Converting Images to TFRecord format

python download_and_convert_data.py \

    --dataset_name=flowers \

    --dataset_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\record

2 从头开始训练 Training a model from scratch

python train_image_classifier.py \

    --train_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\train_log \

    --dataset_name=flowers \

    --dataset_split_name=train \

    --dataset_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\record \

    --model_name=inception_v3

3 预训练模型调优 Fine-tuning a model from an existing checkpoint

python train_image_classifier.py \

    --train_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\train_log \

    --dataset_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\record \

    --dataset_name=flowers \

    --dataset_split_name=train \

    --model_name=inception_v3 \

    --checkpoint_path=H:\pythoncode\Tensorflow\ImageClassify\pretrained\inception_v3.ckpt \

    --checkpoint_exclude_scopes=InceptionV3/Logits,InceptionV3/AuxLogits \

    --trainable_scopes=InceptionV3/Logits,InceptionV3/AuxLogits

4 TensorBoard

tensorboard --logdir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\train_log

5 模型能力评估 Evaluating performance of a model

python eval_image_classifier.py \

  --checkpoint_path=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\train_log \

  --eval_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\validation_log \

  --dataset_name=flowers \

  --dataset_split_name=validation \

  --dataset_dir=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\record \

  --model_name=inception_v3

6 导出推理图 Exporting the Inference Graph

python export_inference_graph.py \

  --alsologtostderr \

  --model_name=inception_v3 \

  --output_file=H:\pythoncode\Tensorflow\ImageClassify\data\flowers\inception_v3_inf_graph.pb
 

  --dataset_name flowers

7 导出权重  Freezing the exported Graph

python freeze_graph.py \

  --input_graph H:\pythoncode\Tensorflow\ImageClassify\data\flowers\inception_v3_inf_graph.pb \

  --input_checkpoint H:\pythoncode\Tensorflow\ImageClassify\data\flowers\train_log\model.ckpt-12780 \

  --input_binary true \

  --output_node_names InceptionV3/Predictions/Reshape_1 \

  --output_graph H:\pythoncode\Tensorflow\ImageClassify\data\flowers\frozen_graph.pb

8 预测 Test

python classify_image_inception_v3.py \

  --model_path H:\pythoncode\Tensorflow\ImageClassify\data\flowers\frozen_graph.pb \

  --label_path H:\pythoncode\Tensorflow\ImageClassify\data\flowers\record\labels.txt \

  --image_file dandelion.jpg

参考资料

[1] https://github.com/tensorflow/models/tree/master/research/slim