Use a pre-trained model to localize and identify multiple objects in a single image or using a Web Camera.



  • Real-Time Web Camera detections!

  • Upload a single image or multiple images and get predictions about the contents of the image. 

  • Javascript only code. 

  • Export results in a JSON file and integrate it easily with your applications. 

  • Web-based, minimal implementation requirenments. It only needs a webserver.  


 Object detection is the task where a model can identify through an image or video stream, information about the objects inside the image and their position.   The current model detects the objects in images using a
single deep neural network. The approach SSD (Single Shot MultiBox Detector), represent the output space of bounding boxes into a set of default boxes over different aspect ratios and scales per feature map location. The prediction generates scores
for the presence of each object. 


The model is trained using the dataset COCO (http://cocodataset.org/#home).


The website works by uploading an image (or group of images) then the model will be called for the prediction of the class, score and bounding boxes. The output will be the original picture with the outlined bounding boxes and the name of the possible
classes with the scores. The user will have the option to export/download the results into a JSON file in case of additional implementations (for example, if the user needs the information to be later stored into a database).



Additionally, using the same model, the option of object prediction via web camera is added to give more options to the user to perform the detections.  This option will require the user to give the browser the permissions for using the webcam. After
the permissions are acquired from the browser, the webcam will start giving the bounding boxes and names of the possible classes in real-time with the web-camera. 




External sources:  The following website uses TensorFlow JS, as well as a pre-trained model for object detection from MobileNet.   It also uses Bootstrap and Jquery.



Limitations



  • The detections correspond to statistical probabilities, hence the accuracy is not 100% guaranteed.

  • The image quality and complexity interferes with the accuracy of the prediction. 

  • The UI is meant to be run using the current modern browsers. The browser should support javascript and enable javascript. 

  • For the web camera function, the user has to give the browser the consent to open and use the web camera. Your website must have enabled https access.  The port 443 needs to be open in order for the browser to ask permission to load the camera
    The class names are only available in English.







User Guide Document
Download
Highest Price
77.17
Lowest Price
39.53
Average Price
57.59
AI Price Forecast
NA
Start to Communicate (If you want to start communication with the seller about this item, you can start here)
Loading…

We use cookies and other technologies to improve your experience on our website and to analyze our traffic. By browsing our website, you consent to our use of cookies and other tracking technologies.

Accept cookies and close this message
Win now