Prepare dataset for action (contact/no-contact) recognition

[caleb-vicente]

How should be an object detection dataset looks like??

• Distribution of train, test, val
• How must be the score be measure?

Some of the task that must be accomplished in this user story:

• [x] Create funtion for object detection that can be imported
• [x] Identify players in contact using helmet tracking data and bounding boxes of whole body
• [x] Join together in peers two bounding boxes of players in contact
• [x] Draw representation of those identifying players under no contact with black bounding boxes as black and the other ones as green
• [x] Apply correction of playes not detected with bounding boxes.
• [ ] Prepare examples of no contact (bounding boxes that are close in distance)

[caleb-vicente]

https://cloud.google.com/vision/automl/object-detection/docs/prepare

• 80% of images are used for training.
• 10% of images are used for hyper-parameter tuning and/or to decide when to stop training.
• 10% of images are used for evaluating the model. These images are not used in training

About the score measure they usually use map/iou:

"The mAP calculation varies in different object detection challenges. According to the COCO 2017 challenge evaluation guidelines, the mAP was calculated by averaging the AP over 80 object classes AND all 10 IoU thresholds from 0.5 to 0.95 with a step size of 0.05" https://www.v7labs.com/blog/mean-average-precision