This repository contains the implementation of our tensor pooling driven contour instance segmentation framework.
To run the proposed framework, please download and install Anaconda. Afterward, please import the ‘environment.yml’ or alternatively install following packages:
Both Linux and Windows OS are supported. To run some utility functions, please install MATLAB R2020a as well along with deep learning, image processing and computer vision toolbox.
The X-ray datasets can be downloaded from the following URLs:
Each dataset contains the ground truths either in mat files, txt files or in xml files. To evaluate the proposed framework, the annotations must be in the mask form. To parse and convert the original box-level annotations to mask-level, we have provided the respective converters in the ‘…\utils’ folder. Please follow the same steps as mentioned below to prepare the training and testing data.
├── trainingDataset
│ ├── original
│ │ └── or_image_1.png
│ │ └── or_image_2.png
│ │ ...
│ │ └── or_image_n.png
│ ├── train_images
│ │ └── tr_image_1.png
│ │ └── tr_image_2.png
│ │ ...
│ │ └── tr_image_n.png
│ ├── train_annotations
│ │ └── tr_image_1.png
│ │ └── tr_image_2.png
│ │ ...
│ │ └── tr_image_n.png
│ ├── val_images
│ │ └── va_image_1.png
│ │ └── va_image_2.png
│ │ ...
│ │ └── va_image_m.png
│ ├── val_annotations
│ │ └── va_image_1.png
│ │ └── va_image_2.png
│ │ ...
│ │ └── va_image_m.png
├── testingDataset
│ ├── original
│ │ └── or_image_1.png
│ │ └── or_image_2.png
│ │ ...
│ │ └── or_image_k.png
│ ├── test_images
│ │ └── te_image_1.png
│ │ └── te_image_2.png
│ │ ...
│ │ └── te_image_k.png
│ ├── test_annotations
│ │ └── te_image_1.png
│ │ └── te_image_2.png
│ │ ...
│ │ └── te_image_k.png
│ ├── segmentation_results
│ │ └── te_image_1.png
│ │ └── te_image_2.png
│ │ ...
│ │ └── te_image_k.png
The additional results of the proposed framework are presented in the '…\results' folder. Please feel free to email us if you require the trained instances.
If you use the proposed framework (or any part of this code in your research), please cite the following paper:
@inproceedings{tensorPooling,
title = {Tensor Pooling Driven Instance Segmentation Framework for Baggage Threat Recognition},
author = {Taimur Hassan and Samet Akcay and Mohammed Bennamoun and Salman Khan and Naoufel Werghi},
note = {Neural Computing and Applications},
year = {2021}
}
If you have any query, please feel free to contact us at: taimur.hassan@ku.ac.ae.