Videology SCAiLX
Scailable supports the Videology SCAiLX camera. If your SCAiLX Zoom Block camera didn't come preinstalled with our Scailable runtime, don't worry, we will take you through the installation step-by-step.
You can quickly install it through our convenient one-line installation process. Simply SSH into the device and run the following command:
sudo bash -ic "$(curl -fsSL https://get.sclbl.net)"
You will see our standard installation feedback, and once the installation is completed, you are good to go!
On a SCAiLX device you should be able to simply browse to
http://scailx6.local:8081 and see our AI manager which enables you to configure your AI pipeline and model on the device.If you are unable to reach the UI of the AI manager at http://scailx6.local:8081, try entering the IP address of the device directly into your browser followed by
:8081.For full AI manager documentation please see: https://docs.scailable.net/the-ai-manager/introduction.
However, for the ScailX device, the "Input" tab will automatically be configured to take as input to your AI models the camera of the device. Alternatively, we provide a small number of test images for easy testing.
Overview of the Input tab after automatic selection of the SCAiLX camera.
We provide a number of demo models specifically configured for the SCAiLX camera in the Videology catalogue. You can find them in our library under the "Videology" heading.
The "Videoloy Models" catalogue.
The current Videology catalogue contains:
- 1.Face locator [cpu]: Locate faces and draw bounding boxes around faces.
- 2.Object detector [cpu]: Locate people, cars, and bikes. Provide bounding boxes and counts.
- 3.Emotion detection [cpu]: Primary emotion detection for the whole image; make sure to be close to the camera.
- 4.Face blurring [cpu]: Using the face locator, we can explicitly blur faces in the resulting output.
- 5.QR code scanning [cpu]: Try to scan a QR code and see the text stored in the image.
Please note that the Videology catalogue is currently only available to users assigned to the Videology organization.
Note that the [cpu] flag behind the models indicates that these models are specified using Floating points and are thus optimized for CPU, not NPU. (see notes below).
You can use the full Scailable platform to import models and deploy them to your SCAiLX device. For example:
- 1.Use ONNX directly to upload any vision model. See our docs: https://docs.scailable.net/for-data-scientist/custom-model-creation
- 2.Use Teachable Machine to train a quick classification model and import it to the Scailable platform: https://docs.scailable.net/for-data-scientist/importing-models/from-teachable-machine
- 3.Use Edge Impulse for model training and deploy directly to the SCAiLX camera: https://docs.scailable.net/for-data-scientist/importing-models/from-edge-impulse
Please note that currently our catalogue models and uploaded models often rely on floating point representations thus limiting the use of the NPU in the NXP iMX8M Plus hardware. Our SCAiLX integration is in active development and we expect the following releases:
- 1.Catalogue NPU release: We will quantize, as much as possible, our catalogue models to support INT8 acceleration on NPU.
- 2.Model upload quantization: We will add automatic quantization of uploaded models (where possible)
- 3.Scailable core runtime update: Currently NPU usage is minimal for specific operators, the currently optimal solution for NPU support within the NXP chip is usage of TFLite models. We are currently adding automatic TFlite conversion support of quantized ONNX models.
Last modified 1mo ago