Buckshot
Fine-tuned CNN for deer detection in camera-trap images, built as a class project. May revisit with finer-grained classification — buck vs. doe, antler presence, or tine counting.
Buckshot
Buckshot is a fine-tuned convolutional neural network trained to detect deer in camera-trap images. The model was developed as a class project, with a written report covering dataset preparation, training methodology, and results.
Camera-trap images present a particular challenge: variable lighting, heavy occlusion, motion blur, and deer that are often partially obscured by vegetation.
Possible Directions
The current model does binary detection — deer or not. Some directions worth exploring:
- Buck vs. doe classification — sex identification from image features
- Antler detection — isolating bucks with visible antlers
- Tine counting — estimating rack size, which is significantly harder and would likely need a dedicated keypoint or segmentation approach