ShearDetect

A defekt detection model for shearographic images. This model is based on a object detection model with faster R-CNN and ResNet-50 approach.

Getting Started

Clone the Code

git clone https://github.com/ILKGit/ShearDetect

Requirements

• Python >3.6
• CUDA 11.3 or higher

Install all the python dependencies using pip

pip install -r requirements.txt

Dataset

A dataset can be find here:

Strucutre of a custom Dataset has to be as following:

|-----train
       |-----annotations
              |-----*.json
       |-----images
              |-----*.tif
|-----validation
       |-----annotations
              |-----*.json
       |-----images
              |-----*.tif
|-----test
       |-----annotations
              |-----*.json
       |-----images
              |-----*.tif

*.json-files contain the following annotations and infos

{
"fileID": "fspecimen_name+image_name",
"Dataset": "specimen_name",
"image": "image_name",
"defect": [[x1, y1, x2, y2],],        #bounding box of defects as list
"specimen": [[x1, y1, x2, y2],].  #bounding box of specimens as list
}

Training / Evaluation

python train_model.py --model=NAME OF YOUR MODEL --epochs=NUMBER OF EPOCHS --save_period=CHECKPOINTS SAVE PERIOD

Detection

python detect_model.py --model=DIR to Model --data=DIR TO DATA --pred=DIR TO SAVE RESULTS