The Artificial Intelligence Device Web Service (AIDWS) is a service that runs on external devices (your own or borrowed), such as a Virtual machine, Jetson or Raspberry Pi. It allows to recognize objects live from newly uploaded photos, recordings or directly from the camera, based on previously trained models.

After clicking Login, you will be redirected to the Web App. If you are logging into the app for the first time, your device status will change to Connected and you will be asked to enter a token in addition to your e-mail and password.

You can copy the token by clicking the Connect button next to the record representing the device you are currently using in the Development tab of the OSAI portal.

After logging into the service, you will find a menu with 4 tabs:

• Cameras
• Live testing
• Results
• Documentation

AIDWS allows you to use your own cameras connected to the device. You can generate images and upload them to a dataset in your account. To use a new camera, click Add camera. We currently support the following camera types:

• Leopard
• Arducam
• Camera CSI

In this section you can configure the camera. The available modes should be selected by loading configurations from the camera.

• Name - camera name
• FPS - number of frames per second
• Resolution - resolution at which the camera will operate

In addition, the Leopard camera has the following options:

• Exposure – responsible for controlling the iris and adjusting the level of incoming light
• Gain – digitally increases brightness
• Auto exposure – automatic attempt to adjust the brightness to the prevailing conditions

After clicking OK, your camera should be visible in the list of enabled devices. From here you can:

• Enter preview mode and generate frames from the image
• Edit camera parameters
• Remove the camera from the list

In the Live View, you can see the camera view and the sections for generating images. The following actions are possible:

• Change the image generation option:
  • By time interval
  • Every X frame
• Select the size to which the generated camera image is to be scaled
• Change the format in which the files are generated
• Choose whether to generate in a new dataset or save to an existing one
• Change the counter of the generated images

Control the process with the Start and Stop buttons.

A device connected to the service automatically displays models that match the computing hardware specified during the device addition process and already converted.

AIDWS displays a list of models available from the OSAI service according to the selected computing hardware. Currently, it is not possible to use a model other than the one from the OSAI service for an application.

On the right side of the Web App main page is a menu that allows you to view test results, change credentials, and log out.

To run the selected model, download it to the device.

Select the model from the list and add the input data. The Web App allows you to test the model with a live video stream or on a saved video file that can be uploaded or selected from those stored on the device. A similar test can be performed on an image.

When everything is ready, the test will automatically start for the loaded data. If you are using data stored on the device, click the Run Model button.

Select Live testing and choose from the configured Cameras to use for your model. If you have one camera configured, go directly to the camera preview and dataset generation section. This allows you to create a dataset from the live recording that has detected classes and annotations.

If you do not have any camera configured, the system will redirect you to the configuration panel.

After running the model on the selected source, you will be redirected to the page with the results.

The image that was previously sent to the server can be found in the Saved Inputs view. You can now run a new (or the same) model on this data.

In the Results section, you will find the input image with the result of the model run added.

In the case of a video file, running the model will take you to a video player page. You can decide if you want to create a dataset based on it in OneStepAI with classes/annotations and view the results in real time. For detection networks, the resulting file will contain the selected recognition areas along with the labels of the objects that have been classified on them.