This is the implementation of manufacturing Object detection and tracking in the manufacturing plants. This model uses Mask RCNN model to do the initial segmentation. Which is based on Feature Pyramid Network(FPN) and a ResNet50 backbone. To give temporal consistency in the detection results, a two-staged detection threshold has been used to boost up weak detections in a frame by referring to objects with high detection scores in neighboring frames.
The repository includes:
- Source code of Mask R-CNN with temporal coherence
- Training code
- Jupyter notebook for detection
- Evaluation file for precision, recall and F1-score
- Install the required dependencies: (for reference see how_to_install.pdf )
- custom.py : this code is used for loading data and training the model
- Training.ipynb: loading the weight and calling the training function
- result_calculation.ipynb: this code is used for detecting objects with or without temporal coherence. This also calculates precision, recall and f1-score of the model.
- mrcnn/visualize_frame_relation_4f.py : this code is used for visualizing the detected objects with mask.
Use the custom.py to set your number of classes and load the dataset (train, val) with annotation. For annotating the images use COCO style image annotation tool, which are available online.
If you use this repository, please cite the following paper:
@article{karim2019region,
title={A Region-Based Deep Learning Algorithm for Detecting and Tracking Objects in Manufacturing Plants},
author={Karim, Muhammad Monjurul and Doell, David and Lingard, Ravon and Yin, Zhaozheng and Leu, Ming C and Qin, Ruwen},
journal={Procedia Manufacturing},
volume={39},
pages={168--177},
year={2019},
publisher={Elsevier}
}