Here's a fun activity whereby the students can understand object detection and training datasets better.

• Divide students in a group of 5 and choose 5 objects.(lets say: leaves, flowers, plants, trees, animals).
• Assign 4 labels to each group. Each group should click 20 photos of each label. (for instance: group 1 get 4 labels: leaves, flowers, trees, animals. Group 1 together should click 80 pictures in total of the 4 labels defined).
• Finally, ask every student of the group to try and click a unique photo of the 5th label (unassigned label)(Now all the kids of group one should try to click a unique photo of plants)
• Train the model on 80 images and let the model predict any 5 images. The group which produces the images which remains undetected or get the lowest sum of confidence wins for clicking the most unique photos.

The outline

Along with the competitiveness and enthusiasm of the young students, they venture into the outdoors in search of finding the perfect shot. The experience of having them escape their comfort zones in an activity that teaches them about how machine learning datasets are trained, will certainly be beneficial to them in understanding how machine learning datasets are trained. This experience might inspire them in the future to return to the this field and work on more complex problems. The bottom line is that, statistically, children learn a lot better when they have a hands-on encounters and they enjoy doing so in the process.