We can now use this trained model to help label more data. Let's go back into the editor and open the 'Automated labelling' drawer.
Object detection and segmentation
Because we just trained an object detection model, the trained model will now appear in the 'Object and segmentation' list of available models.
Once we select it we will get a new list of parameters we can set to run inference with. The main parameters we need to consider are the 'Confidence' and 'Detection range'. The confidence is the minimum model confidence required. By setting the confidence to a certain amount, the inference will not return any predictions below that specified confidence level. The detection range simply states which frames the inference will run over.
We can expand 'Advanced settings' and additionally set whether inference is to be run on CPU or GPU (GPU recommended) and the 'Intersection over union' threshold, which deletes boxes and polygons that have over that threshold overlap with other boxes and polygons.
Run inference and review results
Now we can run the inference and see the predictions return and get rendered in the editor. Note that we can also see the confidence of the predictions above the object boundary. We can now use our 'Bulk label operations' button to delete boxes below a certain threshold or in a range of specific frames if we see the prediction has not done a good enough job.
We can run inference with classification models as well. Select the model you want to run from the list of models under 'Classification' in the 'Automated labelling' drawer.
Classification inference will run return results for frames that have not already been classified under that specific classification 'question'.
Note that for running inference on nested classifications, the results will only return if the 'parent' classification has been made. So for instance, let's say we have a classification structure that asks 'Is there a car in the frame' and if the answer is yes a nested question of 'Is the car red'. If you run a model that trains on 'Is the car red', it will only return inference results for frames that you have classified as having a car in the frame. This ensures that conditional relationships are always enforced, even with automated labelling.