Quickstart

Things to remember:

An example of this for the 2023 season might be:

Mount your camera to the robot in its final position
Connect the USB cable to a laptop or other device
Open up any video recording software (such as camera on windows)
Select the camera you will be using on the robot
Record the object from different angles, distances, and lighting conditions by moving the robot around for 30 seconds, while still keeping the object in frame
Repeat step 5 for each object you want to detect

./run.sh

Select Manual Label from the menu
Copy the path to the video you want to label and paste it into the prompt (Ex. train_data/2023.mp4)
Enter the amount of frames you want to skip between each frame you label (useful only if your video is very long) If you want to label every frame, enter 0
Press the Backtick key and any letter on your keyboard to set a label for that frame
Click on the upper left corner of your object to set the top left corner of the bounding box and click on the lower right corner of your object to set the bottom right corner of the bounding box (the program should look something like this)
Press l to move to the next frame
Repeat steps 4-6 until you have labeled all frames
Press q to quit the program
Repeat steps 1-8 for each video you want to label

Select Semi Automatic Label from the menu
Select Find Bounding (FIRST) from the submenu
Copy the path to the video you want to label and paste it into the prompt (Ex. train_data/2023.mp4)
Press any letter on your keyboard to set a label for that frame
Click on the upper left corner of your object to set the top left corner of the bounding box and click on the lower right corner of your object to set the bottom right corner of the bounding box (the program should look something like this)
Press q to quit the program
Select Semi Automatic Label from the menu
Select Track Bounding (SECOND) from the submenu
Copy the path to the video you want to label and paste it into the prompt (Ex. train_data/2023.mp4)
Check to make sure the bounding box looks correct
Press y if the bounding box is correct, or n if it is not
The program will then track the object in the video and label each frame. If the bounding box ever looks incorrect, press the spacebar to pause the video. Use the dots to adjust the bounding box. Then press spacebar again to continue the video
Press q to quit the program manually, or wait for the video to finish and it will quit automatically
Repeat steps 1-13 for each video you want to label

Select Rename Labels from the menu
Enter the path of the directory containing the raw label files and images (Ex. train_data/2023_TrackerCSRT_1.000000)
Enter the old letter used to represent the label (Ex. p)
Enter the new string you want to use to represent the label (Ex. Paw)
Repeat steps 1-4 for each label you want to rename

Select Convert Labels from the menu
Enter the path of the directory containing the training data (Ex. train_data)
Enter the number of files you want to split the training data into (Ex. 1) (This is useful if you have a lot of training data and want to train on multiple computers at once)
Enter the percentage of the training data you want to use for validation (Ex. 0.3 for 30%) (This is so you can test your model on data it has never seen before)

Select Train Model from the menu and the training will automatically start
Wait for the training to finish. This may take a while depending on the amount of training data you have and the power of your GPU
To view the training progress, open up a web browser and go to http://localhost:6006/
Scroll down and expand the loss section
The lower the loss, the better the model is performing. If the loss is not decreasing, you may need to add more training data or train for longer

Provide feedback