aykhara / tensorflow-sushi-detection

Sushi Detection model with Tensorflow

This is just an example for demo.
Last updated: 3/28/2019 with TensorFlow v1.12

Before you start....

Setting Up DLVM on Azure (Ubuntu)

Please following the Quickstarts.

Tensorflow Object Detection

Basically following the installation.

1. Clone the repo - Tensorflow models

https://github.com/tensorflow/models.git
Any SSH and telnet clients are fine!

2. Install Python, tensorflow-gpu, cuDNN, CUDA - specific version

Here is the version list.

$python --version
Should be "Python 3.5.5 :: Anaconda custom (64-bit)" pre-installed on DLVM.

You can also check nvidia driver version.
$cat /proc/driver/nvidia/version

$pip install tensorflow-gpu==1.12.0
$cat /usr/include/cudnn.h | grep CUDNN_MAJOR -A 2
Will show like "#define CUDNN_MAJOR 7 define CUDNN_MINOR 2 define CUDNN_PATCHLEVEL 1" cuDNN version 7.0 should be pre-installed.
$>nvcc --version
Will show like "Cuda compilation tools, release 9.0, V9.0.176" CUDA version 9.0 should be pre-installed.
$nvidia-smi
Check GPU Card info.

3. Install libraries

$pip install --user Cython
$pip install --user contextlib2
$pip install --user pillow
$pip install --user lxml
$pip install --user jupyter
$pip install --user matplotlib

4. Install COCO API

From models/research
$git clone https://github.com/cocodataset/cocoapi.git
$cd cocoapi/PythonAPI
$make
$cp -r pycocotools <path_to_tensorflow>/models/research/

5. Protobuf Compilation

From tensorflow/models/research/
$wget -O protobuf.zip https://github.com/google/protobuf/releases/download/v3.0.0/protoc-3.0.0-linux-x86_64.zip
$unzip protobuf.zip

6. Add Libraries to PYTHONPATH

From tensorflow/models/research/
$export PYTHONPATH=$PYTHONPATH:pwd:pwd/slim

7. Testing the Installation

From tensorflow/models/research/
$python object_detection/builders/model_builder_test.py

If showing "OK" that means you successfully completed the installation!

8. Access to Jupyter Notebook

$jupyter notebook --ip=0.0.0.0 --port=8888
Copy the URL like "http://:8888/?token=*********"

Then go to Azure portal and add inbound port rule "8888"

Go to browser and paste "http://(DNS name - copy from Azure Portal):8888/?token=***********"

Now ready to code with tensorflow!

---

A. Annotate your own dataset

Annotate dataset

VoTT 1.5

Export to Tensorflow Pascal VOC format

You will get the four below.
・Annotations folder
・ImageSets folder
・JPEGImages folder
・pascal_label_map.pbtxt file

B. Train and Test data preparation

Convert XML to CSV

$python xml_to_csv.py
You need to split the csv file into two: train.csv and test.csv before the next step.

Generate tf-record file (Train)

$python generate_tfrecord.py --csv_input=train_labels.csv --output_path=train.record --label_map_path=pascal_label_map.pbtxt

Generate tf-record file (Test)

$python generate_tfrecord.py --csv_input=test_labels.csv --output_path=test.record --label_map_path=pascal_label_map.pbtxt

C. Train and Evaluate model

Train model

$python train.py --logtostderr --pipeline_config_path=sushi_resnet_101.config --train_dir=checkpoints_sushi

Evaluate model

$python eval.py --logtostderr --pipeline_config_path=sushi_resnet_101.config --checkpoint_dir=checkpoints_sushi --eval_dir=checkpoint_sushi

Tensorboard

$tensorboard --logdir=checkpoints_sushi --port 6008