teleportHQ Vision API is a computer vision API specifically trained for detecting atomic UI elements in pictures of hand-drawn wireframes
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Alt text

Vision API

alt text teleportHQ Vision API is a computer vision API specifically trained for detecting atomic UI elements in pictures of hand-drawn wireframes (as seen in the picture above). It uses an architecture based on Resnet101 for extracting features and Faster R-CNN for bounding-box proposals.

The machine learning model was built and trained using TensorFlow.

List of elements it can distinguish: paragraph, label, header, button, checkbox, radiobutto, rating, toggle, dropdown, listbox, textarea, textinput, datepicker, stepperinput, slider, progressbar, image, video.

The API is currently in closed alpha, but feel free to contact us if you want early access.

Using the Vision API


Send all requests to the API endpoint:

Request header

Make sure to add a Content-Type key with the value application/json and a Teleport-Token key with the key provided by us.

Request body

The body of the request is a json with two keys: image and threshold.

  • image is a required string parameter that denotes the direct url to a publicly available jpg or png image.
  • threshold is an optional parameter. Default value is 0.1. The detection model outputs a confidence score for each detection (between 0 and 1) and won't include in the response detections with confidence lower than this threshold.

Request body example:

    "image": "", 
    "threshold": 0.5

Request example

curl \
  -X POST \
  -H 'Content-Type: application/json' \
  -H 'Teleport-Token: your_token' \
  -d '{ 
    "image": "",
    "threshold": 0.5 


If your request is a valid one, you will recieve back a json with the following structure:

        "box": [y, x, height, width],
        "detectionClass": numeric_label,
        "detectionString": string_label,
        "score": confidence_rating

The json contains a list of objects, each one of this objects corresponding to a detected atomic UI element in the image sent in the request. All of the keys will appear in all of the objects in your response array.

  • box contains the coordinates of the bounding box surrounding the detected element. x and y are the coordinates of the top left corner of the box and width and height are self explanatory. All coordinates are normalized between [0, 1] where (0,0) is the top left corner of your image and (1, 1) is the bottom right corner. In other words, if you want to get the pixel coordinates you have to multiply x and width with the width of your image and y and height with the height of your image.
  • detectionClass is the numeric class of the detection.
  • detectionString is the human-readable label of the detection.
  • score represents how confident the algorithm is that the predicted object is a correct / valid one. It takes values between [0, 1], where 1 represents a 100% confidence in its detection.

The detectionClass to detectionString mapping is done according to this dictionary:

    1: "paragraph",
    2: "label",
    3: "header",
    4: "button",
    5: "checkbox",
    6: "radiobutton",
    7: "rating",
    8: "toggle",
    9: "dropdown",
    10: "listbox",
    11: "textarea",
    12: "textinput",
    13: "datepicker",
    14: "stepperinput",
    15: "slider",
    16: "progressbar",
    17: "image",
    18: "video"

Response example

Full response here.

        "box": [
        "detectionClass": 17,
        "detectionString": "image",
        "score": 0.999999
        "box": [
        "detectionClass": 4,
        "detectionString": "button",
        "score": 0.989777
        "box": [
        "detectionClass": 2,
        "detectionString": "label",
        "score": 0.98929

How do I get a Teleport-Token?

If you are interested in using this API, feel free to get in touch with us via the following form.