Overview

Center for Artificial Intelligence (CAI) is challenging all developers to build an algorithm that automatically detects objects live. The training dataset is a walking video through cities. The algorithm must detect many things in the video and show bounding boxes with annotation. Computers are indeed getting better at vision, however, in a few critical scenarios, they cannot match human perception. For example, what do you see when you look at the photo?

Most of us would answer, “People are walking in a hallway, there are vending machines and lookers, and there is a noticeboard, the place is a university”. Nowadays, computers can give similar answers. To make it challenging, instead of static photos, let’s make it dynamic where things change based on the scene. CAI wants to push the capabilities of computer vision and AI. Providing a large training set will stimulate research and development into more sophisticated real-time detection and description models that exceed current cutting-edge performance.

Real-time object detection Track:

Object detection is the core of computer vision. Many computer vision applications are built on object detection, such as robotics, self-driving cars, surveillance, and others. As there are many AI models for such applications, real-time object detection is still challenged in matching the precision of human perception.

To push advances in analyzing and processing videos, CAI will provide a video of a walking street in different cities that will be used for training and validation. In the challenge, you are asked to build the best-performing model for real-time object detection and to achieve the best balance between three crucial factors (speed, accuracy, and memory)

Speed: How quickly can your model identify and localize objects within a live video stream? Measured in milliseconds (ms).

Accuracy: How precisely can your model distinguish and classify objects? Measured by the percentage of correct detections. In addition to specificity, sensitivity, precision, and F1.

Memory: How efficiently does your model utilize computing resources? Measured in megabytes (MB).

Important Dates: