Dataset Generation for Flow-Assisted Diffusion Channel in Molecular Communications
1. Summary
We are looking for motivated student(s) to form a dataset of micro-scale beads from microscopic videos and test it with state-of-the-art techniques of multi-object detection, instance segmentation, and, if time permits, multi-object tracking (MOT).
The main challenge here is that a continuous flow making moving micro-scale beads look blurry, very fast to detect, and the existence of false-negative candidates.
A publication of how these videos are being captured in a laboratory environment is at [1] and some examples of frames are presented in Section 4.
2. Expected Outcomes
We have only one class of object (C=1) to detect, beads; but we have many of them in a frame, hence this yields to a multi-object detection problem. So far, we only used yellow and red-colored beads. The variability of sharpness of videos is high, which makes the problem challenging. We expect annotations of color and a boolean value of blurry, and more, if you want.
There are many tools for dataset generation, some are labelme [2] and VGG Image Annotator [3]. You are free to choose any convenient one.
We expect the dataset, related codes, and others to be at https://github.com/nanonetworking/p1001-flowing-beads-dataset. We can provide you our own GPUs within our limitations.
A comprehensive list of object detectors is at [4] and [5]. More can be found with literature research.
3. Publication
We aim to publish the dataset to the public via a conference paper or such. Student(s) completing this project will be an author of it. Therefore, we expect a commitment to the project.
4. Examples
References
- M Gorkem Durmaz, Abdurrahman Dilmac, Berk Camli, Elif Gencturk, Z Cansu Canbek Ozdil, Ali Emre Pusane, Arda Deniz Yalcinkaya, Kutlu Ulgen, and Tuna Tugcu. Preliminary studies on flow-assisted propagation of fluorescent microbeads in microfluidic channels for molecular communication systems. International Conference on Bio-inspired Information and Communication Technologies, pages 294–302. Springer, 2020.
- Kentaro Wada. labelme: Image Polygonal Annotation with Python. https://github.com/wkentaro/labelme, 2016.
- Dutta, A. Gupta, and A. Zissermann. VGG image annotator (VIA) http://www.robots.ox.ac.uk/ vgg/software/via/, 2016.
- Object detection. https://handong1587.github.io/deep_learning/2015/10/09/object-39detectio... Accessed: 2021-03-05.40
- Awesome object detection. https://github.com/amusi/awesome-object-41detection. Accessed: 2021-03-05.