Object detection on Carbon Fibres

References

• https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/

To use google colab

To get the code

• get the source code folder from gitlab !git clone -b dev https://git.rwth-aachen.de/athith.boloor/defect_detection_carbon_fibre.git

Populate the pre trained model folder

• run the code !wget http://download.tensorflow.org/models/object_detection/ssd_inception_v2_coco_2018_01_28.tar.gz
• extract the file using the command !tar xvzf ssd_inception_v2_coco_2018_01_28.tar.gz -C defect_detection_carbon_fibre/pre-trained-model/
• Cut/Move the content of this folder outside(one level above)mv defect_detection_carbon_fibre/pre-trained-model/ssd_inception_v2_coco_2018_01_28/* defect_detection_carbon_fibre/pre-trained-model/

Populate the data

• get the data stored in zip format from Amazon s3 !wget https://thesis-master.s3-eu-west-1.amazonaws.com/project/images.zip
• unzip to the folder images using the following command !unzip images.zip -d defect_detection_carbon_fibre/images/

Train and save the model

• Change the directory from /content to /content/defect_detection_carbon_fibre by executing this code import os os.chdir(os.getcwd()+"/defect_detection_carbon_fibre/")
• execute the command !python train.py
• first download the saved model (To carry out the inference later)and convert it into the zip format and save it in AmazonS3

Inference

Retrieve the trained model

• retrieve the saved model from Amazon s3 using the command !wget https://thesis-master.s3-eu-west-1.amazonaws.com/project/trained_model.zip
• unzip the model into the directory defect_detection_carbon_fibre/training/ using !unzip trained_model.zip -d defect_detection_carbon_fibre/training/

Exporting a trained inference graph

• change working directory to /defect_detection_carbon_fibre/ using the command import os andos.chdir(os.getcwd()+"/defect_detection_carbon_fibre/")
• run the following command to generate trained inference graph in the directorytrained-inference-graphs/output_inference_graph_v1.pb using the following command !python export_inference_graph.py --input_type image_tensor --pipeline_config_path training/ssd_inception_v2_coco.config --trained_checkpoint_prefix training/model.ckpt-20000 --output_directory trained-inference-graphs/output_inference_graph_v1.pb

Predict on test data

• run !python infer.py to make predictions on the test images.