3D Object Analysis - Meshes/Spots

ATL (Aivia That Learns) and the Auto Icon

3D Object Analysis - Meshes features AI-powered automatic parameter selection (sometimes referred to as Aivia That Learns, or ATL). This is an internal model, trained by human experts, that given the selected input image channel, predicts the parameters and populates the Recipe Console with predicted parameters.

There are two ways that ATL is activated for 3D Object Analysis:

• By clicking the 3D Object Analysis purple tile in the search bar (see below). ATL is activated and predicts all the parameters of the recipe automatically. This mode had the advantage that you can try to preview the result in the selected ROI, with just two clicks (first the purple tile, then on Preview, the Magnifier lens besides the description of the parameters).

• Click on the Auto icon (see figure to the right) in the recipe header to initiate the parameter prediction.

Quick advices for using automatic selection

1. Image Smoothing Filter Size and Mesh Smoothing Factor are parameters that affect the smoothness of the final mesh. ATL might not predict these parameters, so you can experiment with them after running the recipe with ATL-generated parameters. Try small increments of the values and check in the preview to achieve nicer visualization of the current segmentation. We recommend using Image Smoothing Filter Size if you observe the image is noisy, and Mesh Smoothing Factor to increase the smoothness the final mesh (less irregularities and peaks).

2. ATL gives you a suggestion for parameters but occasionally may yield less than perfect results due to noise in data, variations on brightness and contrast etc. You might use ATL-generated values as starting point, then check in the Recipe console about the different parameters that can be adjusted to fine-tune the results. There are three possible cases when the prediction might yield less than ideal results:

• Only a few objects are displayed: An image might be very heterogeneous (has objects of multiple sizes, both very big and very small or has low SNR). In this case, some objects might be shown while others are missing. A good starting place is first to modify the Min Edge Intensity (reduce it to around 30%), then the Object Radius (min radius value can be smaller and max radius value can be bigger), and finally the Min Edge to Center Distance (only if applies because Partition was selected, as a rule of thumb, it must be the average size of the object of interest). Other parameters might need further adjustment to refine the resultant segmentation.

• Too many objects are displayed: Similar to the option above, modify first the Min Edge Intensity (increase it to around 10% each time until satisfied), then the Min Edge to Center Distance (if applies), and if you want to crop out small objects, the Object Radius (min value). Other parameters might need further adjustment to refine the resultant segmentation.

• No results are shown in preview: Start modifying the Min Edge Intensity by putting half the value shown in the prediction. If it does not work, keep reducing the value. If you arrive to 0 in intensity, check the Min and Max value of the Object radius.

3D Object Analysis Active Tile

Another way to initiate parameter estimation for 3D Object Analysis - Meshes is through the 3D Object Analysis Active Tile, which is accessed from the search bar at the top of the Aivia window.

The 3D Object Analysis Active Tile also adjusts the Aivia layout to one that is recommended for an object analysis workflow and creates and/or selects a region of interest (ROI) that may be useful for previewing and evaluating the parameters.

Detection

Partition