Object Detection in video

[caleb-vicente]

Investigate different versions of YOLO (v3, v4, v5). Some of them are faster, other ones more accurate.

There are probably some object detection models used specifically for soccer or american football (Example: https://medium.com/analytics-vidhya/player-detection-using-deep-learning-492122c3bf9)

https://towardsdatascience.com/object-detection-using-yolov3-and-opencv-19ee0792a420

[caleb-vicente]

thre are some models pretrained to detect american football players. See how to use them in python without calling an API: https://universe.roboflow.com/bronkscottema/football-player-detection/dataset/3

In this video they show you how to finetune a yolo model: https://www.youtube.com/watch?v=5nsmXLyDaU4

yolov7 breakdown: https://blog.roboflow.com/yolov7-breakdown/

[caleb-vicente]

Fight.v1-fight_set.yolokeras.zip

[caleb-vicente]

Libraries to try YoloV7:

• https://huggingface.co/spaces/kadirnar/yolov7
• use original library: https://inside-machinelearning.com/en/use-yolov7/
• finetune yolo: https://blog.paperspace.com/yolov7/
• finetune yolov7: https://learnopencv.com/fine-tuning-yolov7-on-custom-dataset/

[caleb-vicente]

Try another types of modules. Fast RCNN parece ser mejor para casos en los que estemos detectando objetos pequeños (https://everitt257.github.io/post/2018/08/10/object_detection.html)

[caleb-vicente]

https://universe.roboflow.com/umbc-linyj/megafootballplayer/browse?queryText=&pageSize=50&startingIndex=0&browseQuery=true