Train a Custom Object Detection Model Using TensorFlow APIs

How to Run

Alternatively, if you want to use your images instead of ones comes with this repo.

Require Python 3.5+ installed.

Fork and clone this repository to your local machine.

https://github.com/Tony607/object_detection_demo

Install required libraries

pip3 install -r requirements.txt

Step 1: Annotate some images

Save some photos with your custom object(s), ideally with jpg extension to ./data/raw directory. (If your objects are simple like ones come with this repo, 20 images can be enough.)
Resize those photo to uniformed size. e.g. (800, 600) with

python resize_images.py --raw-dir ./data/raw --save-dir ./data/images --ext jpg --target-size "(800, 600)"

Resized images locate in ./data/images/

Train/test split those files into two directories, ./data/images/train and ./data/images/test
Annotate resized images with labelImg, generate xml files inside ./data/images/train and ./data/images/test folders.

Tips: use shortcuts (w: draw box, d: next file, a: previous file, etc.) to accelerate the annotation.

Commit and push your annotated images and xml files (./data/images/train and ./data/images/test) to your forked repository.

Step 2: Open Colab notebook

Replace the repository's url to yours and run it.

How to run inference on frozen TensorFlow graph

Requirements:

frozen_inference_graph.pb Frozen TensorFlow object detection model downloaded from Colab after training.
label_map.pbtxt File used to map correct name for predicted class index downloaded from Colab after training.

You can also opt to download my copy of those files from the GitHub Release page.

Run the following Jupyter notebook locally.

local_inference_test.ipynb

How to run TensorFlow object detection model faster with Intel Graphics | DLology Blog

How to deploy the trained custom object detection model with OpenVINO

Requirements:

Frozen TensorFlow object detection model. i.e. frozen_inference_graph.pb downloaded from Colab after training.
The modified pipeline config file used for training. Also downloaded from Colab after training.

You can also opt to download my copy of those files from the GitHub Release page.

Run the following Jupyter notebook locally and follow the instructions in side.

deploy/openvino_convert_tf_object_detection.ipynb

Run the benchmark

Examples

Benchmark SSD mobileNet V2 on GPU with FP16 quantized weights.

cd ./deploy
python openvino_inference_benchmark.py\
     --model-dir ./models/ssd_mobilenet_v2_custom_trained/FP16\
     --device GPU\
     --data-type FP16\
     --img ../test/15.jpg

TensorFlow benchmark on cpu

python local_inference_test.py\
     --model ./models/frozen_inference_graph.pb\
     --img ./test/15.jpg\
     --cpu

jnbli/tensorflow-object-detection