FRC Robot Detection & Analysis

CLI for training and testing Robot Detection and Re-Identification models

Demo

Use

To use the project you can download the release, unzip the file, then run the executable as a CLI, use main.exe --help to get started.

When training a model with robot make sure your data is in the correct format (e.g. yolo26, yolo11, yolo8, etc.) When traing a model with reid make sure your data is fomatted as:

dataset/
├── train/
│   ├── robot_001/
│   │   ├── 0001.png
│   │   ├── 0002.png
│   │   ├── 0003.png
│   │   └── ...
│   │
│   ├── robot_002/
│   │   ├── 0001.png
│   │   ├── 0002.png
│   │   └── ...
│   │
│   └── ...
│
└── val/
    ├── robot_001/
    │   ├── 0001.png
    │   └── ...
    │
    ├── robot_002/
    │   ├── 0001.png
    │   └── ...
    │
    └── ...

If you're using the robot framework, in the cli you should not set your dataset as that directory but rather a yaml file formatted as:

path: /relative/path/to/dataset

train: train/images
val: valid/images
test: test/images

nc: 1
names:
  - object_name

Currently, there is no validation or testing for ReID models.

Notes

The full model I use in my demos videos is large and computationally intensive. To get to that point would require a nicer GPU and an upwards of an hour of training time. For faster training, turn down the epochs and use a model like yolo26s.pt

This project is intended for analysis and demonstration purposes only.

Future Work

• Improved analytics for robot performance and movement patterns.
• Web interface for interactive video review.
• Robot to field translation