-
Take a look at existing models
-
Apply this on cifar10 dataset.
An informal term we often use for the layer just before the final output layer that actually does the classification.
- This penultimate layer has been trained to output a set of values that's good enough for the classifier to user to distinguish between all the classes it's been asked to recognize.
git clone https://github.com/googlecodelabs/tensorflow-for-poets-2
cd tensorflow-for-poets-2
get some example images
curl http://download.tensorflow.org/example_images/flower_photos.tgz \
| tar xz -C tf_files
After setup, the folder structure should look like this:
├── android
│ ├── tflite
│ └── tfmobile
├── CONTRIBUTING.md
├── LICENSE
├── README.md
├── scripts
│ ├── count_ops.py
│ ├── evaluate.py
│ ├── graph_pb2tb.py
│ ├── __init__.py
│ ├── label_image.py
│ ├── quantize_graph.py
│ ├── retrain.py
│ └── show_image.py
└── tf_files
└── flower_photos
The retrain script can retrain either:
- Inception V3 model
- MobileNet
In this example, we will choosse Inception V3
-
Input image resolution: 128,160, 192 or 224
-
The relative size of the model as a fraction of the largest MobileNet: 1.0, 0.75, 0.5 or 0.25.
we can set 2 shell variables as follows :
IMAGE_SIZE=224
ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"
python -m scripts.retrain \
--bottleneck_dir=tf_files/bottlenecks \
--how_many_training_steps=500 \
--model_dir=tf_files/models/ \
--summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \
--output_graph=tf_files/retrained_graph.pb \
--output_labels=tf_files/retrained_labels.txt \
--architecture="${ARCHITECTURE}" \
--image_dir=tf_files/flower_photos
The retraining script writes data to the following two files:
-
tf_files/retrained_graph.pb, which contains a version of the selected network with a final layer retrained on your categories.
-
tf_files/retrained_labels.txt, which is a text file containing labels.
Here are command to use the model.
python -m scripts.label_image \
--graph=tf_files/retrained_graph.pb \
--image=tf_files/flower_photos/daisy/21652746_cc379e0eea_m.jpg