First of all image pre-processing has been done by making use of Opencv(Canny Edge and Auto canny, Clahe, various type of Morphological operation like Erosion , Dilation , Medium blurr for feature extraction in image).
Then annotate the images for better understanding of crack area & locating ground truth in image. Took annotated image reading in xml file the converted it to CSV file & for applying it to tensorflow vector converted it to .rec file.
Next step is to use pre-processed image & converted file and applying it to built in SSD-Inception model for having good accuracy & took inference of images.
Calculated the mAP, Precision, Recall, Classification Accuracy