Bounding Box
A standard way of describing object position in object detection. A rectangle with coordinates represents the object region, usually comprising the top-left corner coordinates together with width and height.
Timeline (last 90 days)
The output of an object detection model usually contains three elements: label, bounding box, and confidence score; the core difference from image classification is that object detection also needs to locate the object's specific position in the image.
Keypoint localization requires a model to give precise coordinates of several semantic parts on a target, with higher spatial precision requirements than bounding boxes.
Many end-to-end parsing models based on VLMs only output text sequences without positional information, which creates an obstacle in scenarios that require locating text in the source document.
A rectangular bounding box on the ERP plane cannot uniquely determine the center position and angular range of a visible person on the sphere.
Bounding boxes use only rectangles to locate targets, with low annotation cost and fast speed, suitable for scenes with little occlusion between targets; segmentation masks outline target contours pixel by pixel, enabling precise differentiation of overlapping targets.
The annotation cost for segmentation masks is typically 3 to 5 times that of bounding boxes.
The model's data extraction is accurate, but the generated bounding boxes are sometimes imprecise.
In native visual analysis, DeepSeek Flash extracts content (such as dimension data) accurately, but bounding boxes frequently exhibit offsets with imprecise positions.
All Facts (8)
The output of an object detection model usually contains three elements: label, bounding box, and confidence score; the core difference from image classification is that object detection also needs to locate the object's specific position in the image.
50%UnverifiedKeypoint localization requires a model to give precise coordinates of several semantic parts on a target, with higher spatial precision requirements than bounding boxes.
50%UnverifiedMany end-to-end parsing models based on VLMs only output text sequences without positional information, which creates an obstacle in scenarios that require locating text in the source document.
50%UnverifiedA rectangular bounding box on the ERP plane cannot uniquely determine the center position and angular range of a visible person on the sphere.
50%UnverifiedBounding boxes use only rectangles to locate targets, with low annotation cost and fast speed, suitable for scenes with little occlusion between targets; segmentation masks outline target contours pixel by pixel, enabling precise differentiation of overlapping targets.
50%UnverifiedThe annotation cost for segmentation masks is typically 3 to 5 times that of bounding boxes.
50%UnverifiedThe model's data extraction is accurate, but the generated bounding boxes are sometimes imprecise.
50%UnverifiedIn native visual analysis, DeepSeek Flash extracts content (such as dimension data) accurately, but bounding boxes frequently exhibit offsets with imprecise positions.
50%