Getting Started with Object Detection

The steps below outline HUNTR's basic workflow.

Make sure you are using a supported web browser. HUNTR currently supports Google Chrome, Mozilla Firefox, and Microsoft Edge.

1. Upload a scene to HUNTR:

   • Click the ADD SCENE button in the upper right-hand corner of the home screen and upload your scene or scenes. Scenes uploaded to HUNTR will be shared between all users on the HUNTR subscription account.

   • You can also add a new folder and add scenes from within a folder.

   

   Add your first scene

   See Input and Output Filetypes for more information on accepted input.

2. Create an AI model:

   • Enter the ORCA HUNTR - MAP view by clicking on a scene. Under AI Model on the left sidebar, click CREATE NEW. Models created in HUNTR will be shared between all users on the HUNTR subscription account for your organization. Models created by one user can be used for detection but not trained or changed by others. The specifications for the model can be downloaded as a JSON file by clicking DEFINITION.

   

   Click CREATE NEW

   • Give your model a name and state the author.
   • Pick from one of the predefined color schemes and add your detection classes. You will need to define at least one class and up to 10 classes are supported.
   • Click CONFIRM when done.

   

   Create your first model

3. Create a training area:

   • First, choose the TRAINING AREA button above the scene.
   • Find an area on the scene you would like to label and click to create a training area.

   

   Create your first training area

4. Label and train:

   • Click the LABEL button that appears on the right side of a training area.

   

   Click to label

   • Label your training area using the toolbar to select classes and tools. See Tips for Best Results for more info.
   • Click START TRAINING located on the right-hand side of the toolbar or on the preview image to the right to begin training and receive a live preview. It may take a few minutes to allocate the cloud GPU.¹

   

   Label and get a preview

   • When you are happy with the labels and prediction preview, you can stop training with the STOP TRAINING button in the upper right-hand corner of the training preview screen or simply go back to the map and repeat steps 3. and 4. It is best to create and label many training areas before moving on to step 5., optimally across several scenes.

5. Run detection on your scene:

   • Click the DETECTION RUN button above the scene to start large-scale detection.
   • Click and drag to choose an area and confirm your selection.
   • The model will now run detection on the selected area and show the progress along with intermediate results. If you are not satisfied with the intermediate results, you can stop the run by clicking STOP DETECTION RUN in the RESULTS dialogue box to the right and go back to labeling and training.

   

   Run detection to find target features

6. View stats and export the results:

   • After a detection run, you can view your data by clicking the DATA tab in the RESULTS dialogue box. Choose from a class segmentation mask, a probability mask, model entropy,² or view detections as vector data with detection centroids, bounding boxes, and polygons.

   

   View data and insights

   • Data from completed runs can be viewed by choosing the run under Previous Detection Runs on the left sidebar.

   • Export and download the data in the formats shown in Input and Export Filetypes.

_{1. Starting training allocates a cloud GPU. Cloud GPUs are started and stopped to conserve resources. If you would like to have an always-on GPU, that can be made available with a change to the subscription, please contact inquiries@blackshark.ai for more information.}

_{2. This shows the amount of entropy associated with the prediction. Clearer borders and lower-entropy areas show a higher degree of certainty. Areas of high entropy are good places to create additional labels.}