Image Sources

Weapon Keypoint Detection

The first model in the length measurement module is the keypoint detection model. The dataset used to train this model comes from the Classification dataset . However, as the scope of the measurement module only applies to long guns, not all typologies are necessary.

Card Detection Dataset

The second model is designed to detect cards in the image, providing a reference object to measure the gun against. The dataset used was generated using multiple card images and backgrounds combined with data transformations.

Quality Verification for the Firearms Dataset

To ensure effective training, the dataset must adhere to some guidelines:

• Each image should contain only one firearm.
• The firearm should be entirely within the frame of the image.
• The right side of the firearm should be visible unless the mechanism is visible from the other side.
• The firearm should not have any accessories or a folding/telescopic stock.
• The image should be sharp enough to see the mechanism.

Labeling

The tool used for labeling is CVAT.

For more information on the dataset organization and format you can check this page