I have made a plastic object detector by employing methods of deep learning. When the same is used on an image it returns the bounding box, class name and confidence score, if found.
I gathered and annotated a dataset of 2000 images and labelled it accordingly in PASCOLVOC3 format
Next I converted the dataset to csv files using xml_to_csv.py
I have made this model using the faster masked RCNN from Tensorflow's Object Detection API, which requires input to be fed as tfrecords which I do by using generate_tfrecord.py
I uploaded the dataset to my google drive and trained the faster masker RCNN on Google collab using the same. I have included the notebook "obj.ipynb" for the same. I have made a predict function in the same notebook. In future I will upload it as a .py file and seperate the train and predict parts into different files.
Dependencies : Tensorflow Object Detection API Pandas PIL