Detection Head

Also known as: Object Detection Head, Box Head

A detection head is the output module of a classical object detector (YOLO, RetinaNet, Faster R-CNN). It predicts box geometry and class scores as tensors, usually followed by non-maximum suppression (NMS). Training uses localization and classification losses on those tensors, not next-token cross-entropy on a JSON string.

Contrast with VLM grounding

A vision-language model that “returns boxes” still emits text (digits and punctuation). There is no detection head. Parsing the string and drawing rectangles is post-processing outside the model.

← Back to AI Glossary