Tips for Successful Object Collection

Goals of Object Collection

To build an accurate model, you need to start with accurate and complete data. When collecting an object with AmniSphere, the goal is to capture the highest possible quantity of quality images. A high-quality collection typically consists of:

• No background noise other than the surface where an object is captured.
• Good contrast between the object and the background.
• Object framing does not contain unnecessary empty space (high object pixel density).

Prior to Object Collection

It is important to prepare an environment that is conducive to collecting accurate data on your objects. Before beginning the collection process for an object, make sure:

• You have already added the object to your library in the appropriate folder.
• The objects you would like to collect are nearby for easy access.
• The background you are using to collect your objects is different from the background you are using to run inference on your objects.
• The background you are using should be as free of noise as possible – remember this is not the same background where you will be running inference, so unnecessary noise will not be an advantage to you here.
• The background you are using should not be a similar color to the object you are collecting. For example, you should not collect a white container on a white table.
• The background you are using should not be reflective. Our app uses camera flashes to simulate different lighting settings and this can result in distorted images.

During Object Collection

Once you’ve prepared an appropriate environment for collection, focus on capturing as much data as possible on your objects.

• Ensure the object takes up as much of your screen as possible without cropping any part of the object.
• Do not use any digital zoom when collecting your objects.
• Hold your device steady to prevent blurry images.