MuhammadLab
AI ToolsBrowser-basedPre-trained modelComputer visionStudent demo

Object Detection

Upload an image and see how a pre-trained AI model detects objects using bounding boxes and confidence scores.

Upload an image

Choose an image for detection

Supported formats: .png, .jpg, .jpeg, .webp.

Preview

Image + overlay

Upload an image to see the preview and bounding boxes.

Results

Detected objects

Confidence is the model’s estimate. Filtering helps you see how thresholds change what is shown.

Model status

Idle — Loading COCO-SSD model...

Confidence threshold: 50%

Lower threshold shows more detections (and more mistakes). Higher threshold shows fewer detections (usually more reliable).

Count

0

No objects shown at this threshold.

Confidence meaning

What does confidence mean?

A confidence score of 92% means the model strongly favours that label, but it does not guarantee correctness. Adjusting the threshold helps you see filtering trade-offs.

Step-by-step pipeline

How object detection works

Object detection finds what objects are present and where they are located.

1

Image input

You upload an image (kept local in the browser).

2

Resize / preprocess

The model expects images in a consistent format and size.

3

Model scans features

The neural network looks for learned visual patterns.

4

Predict object class

It predicts categories from the COCO dataset.

5

Generate bounding box

For each object, it estimates a rectangle location.

6

Assign confidence

Each prediction includes a confidence score.

7

Display detections

The UI draws boxes and shows the filtered results.

Detection table

Bounding box coordinates

COCO-SSD outputs \((x, y, width, height)\) boxes in image pixel coordinates.

ObjectConfidenceXYWidthHeight
No detections to display.

How the model makes predictions

Pre-trained visual patterns

The model has already been trained on many labelled images. It does not “understand” the image like a human.

It detects visual patterns learned during training and predicts object categories from the COCO dataset.

Each prediction includes a class label, a bounding box, and a confidence score.

Understanding the output

Labels, boxes, thresholds, and mistakes

Class label: the predicted object name.

Confidence score: how sure the model is.

Bounding box: the rectangle around the object.

Threshold: the minimum confidence needed to show a result.

False positive: when the model detects something incorrectly.

False negative: when the model misses an object.

Limitations and ethics

Use detection responsibly

  • The model can make mistakes.
  • It may perform worse on unusual images, poor lighting, occlusion, or objects outside its training dataset.
  • Confidence is not the same as truth.
  • Object detection should be used carefully in surveillance, identity-related, or high-stakes contexts.
  • This demo is for learning, not clinical, forensic, or safety-critical decision-making.

Student learning outcomes

What you will learn

  • Understand what object detection means.
  • Understand the difference between classification and detection.
  • Interpret confidence scores.
  • Understand bounding boxes.
  • Understand why AI models can make mistakes.
  • Learn how pre-trained models can run in the browser.

Privacy note

This demo runs COCO-SSD in the browser and does not upload your image to a server.