With recent advancements in Artificial Intelligence and deep learning, capabilities for machines to recognise objects and patterns have increased significantly. There has also been an increased interest in detecting and classifying human activities and actions using computer vision to help us in our daily lives.

In this challenge, teams will compete to build and train AI models to perform computer vision, by classifying images according to their defined categories.

Data (Pre-Competition)
During the pre-competition briefing, labelled image datasets will be released to the teams. Teams are encouraged to augment and expand the datasets to improve the generalisation of their solution.

Data (Competition Day)
On the competition day, a dataset with new but related labels will be released. Teams will be evaluated based on how their solution performs for both old and new datasets.

Teams will be given access to an online platform installed with a deep learning toolbox. The online platform will be similar to a Jupyter notebook or Google Colab with standard deep learning environment libraries installed.

During the pre-competition briefing and workshop, teams will be provided with scripts to load/pre-process data and solutions that achieve a minimal baseline performance. Teams are expected to improve upon the baseline solution.

Judging Criteria
Key judging criteria include the team’s model performance, creativity in problem solving and ability to clearly explain and present their solution.

Model performance will be evaluated on a test dataset against popular metrics. The creativity of the team in processing datasets, presenting new model architectures and applying innovation will be considered. Teams are encouraged to explain their data preparation process, modelling techniques, choice of model and innovations.

 TIL | Programme | Registration | FAQs | Contact Us