Frequently Asked Questions
- Anyone, corporate teams, individuals, startups, veterans, and academia with a promising idea towards plastic menace and the tenacity to implement their concept can apply.
- It is an online hackathon, and you can participate from anywhere.
- The hackers must register in a team of a maximum of four members (no minimum). While registering, all members must share their profiles.
- All participants/team are expected to submit their final ideas in a PowerPoint Presentation (PPT) format along with code. You can download the standard PPT template here. Refer to the submission rules mentioned.
- The Idea (phase I) of this hackathon will be open from the 9th of June 2023 till the 9th of July 2023. The submission window will close by 11.59 PM on the 9th of July 2023.
- This hackathon’s prototype (phase II) will be open to only short-listed submissions (watch out for the website and corresponding mail for intimation). The timeline to submit the final prototype with code and slide deck is 11.59 PM on 30th July 2023.
- The submitted solution must be original; if found plagiarised, the submission will be disqualified.
- The teams may use open-source libraries and other freely available systems like labeling tools, GPU with Google Collab, Kaggle, Data Bricks, etc.
- If any of the Teams would want GPU support from the University, you may approach us.
- For labeling the images, you may use free tools like CVAT.
- The intellectual property of your submissions belongs to Kyndryl.
- AI Developers, AI Enthusiasts, Data Scientists, ML Engineers, Programmers, freelancers, students etc, can participate.
- Participants must have experience with object detection models and frameworks, such as YOLO, TensorFlow Object Detection API, or OpenCV.
- Participants must have a good understanding of image processing techniques and deep learning algorithms.
- Participants should demonstrate their ability to implement and fine-tune object detection models for real-world applications.
- Participants must be familiar with data annotation and cleaning techniques.
- Participants should have experience in programming languages such as Python.
- They must be able to work collaboratively in teams and communicate effectively.
- They must be able to submit a detailed report, codes, and relevant files.
- Participants should have a strong passion for creating innovative solutions to tackle the global plastic waste problem.
AI Solutions to Tackle the Plastic Menace: The Hackathon is on detecting plastic in water bodies like rivers of select locales. The participants of the Hackathon would have to detect, classify, and segment the plastics in the given dataset and propose a viable and architecturally sound AI solution to curtail the plastic menace choking the water bodies like rivers, preventing them to be taken into the sea and contributing to some of the garbage patches like Pacific Patch. The solutions must have actionable insights where the stakeholders, like the government and local support groups, can take preventive or proactive measures.
The problem statement of this hackathon is to develop a reliable and efficient AI-based object detection algorithm using images captured via drones to detect plastic waste in rivers and other water bodies and, in turn, to reduce the negative impact of plastic pollution on the environment and human health.
The metadata of the UAV images includes geo-location information. The images are taken at an elevation approximately 10m above the water level.
Guidelines for Building an AI Model on Object Detection
- Participants are required to label the data on their local system before training. You’ll need to annotate the images using an annotation tool such as Labelimg/CVAT.
- You may use any open GPU systems like Kaggle/Google Colab/Data Bricks/Cloud platforms for training the models.
- Train and test datasets are provided, and it is important to use the training set exclusively for training the model, while the test set should be used for testing the model. Register here to download the data set.
- Organize your object detection algorithm’s folder structure, including folders for training, testing, and any data configuration files that may be necessary while submitting.
- Maintain clean coding practices, including proper inline comments, which must be followed.
- The code should effectively detect plastic waste and provide its geo-location as a URL link.
- The model’s mean average precision, commonly referred to as mAP, needs to be a minimum of 65%.
- The estimated effort to tackle this data challenge is around 40 hours.
Instructions for Submission:
- Submit the complete model with a folder structure that adheres to the instructions
- Include Jupyter Notebook or Python file used for training, as well as a requirements.txt file, also in ONNX format
- Zip and submit all in the given link.
- Submit the presentation deck as per the template given.
- Submit the video explaining the approach and solution.