Once you have trained a model version, in addition to looking at the evaluation metrics and estimating how good in general your model is, another challenge is to understand the strengths and weaknesses of your model (on which kind of images it is good, on which kind of images it needs to improve).
To do so, you can select the model you are interested in as Current in the model listing page, by clicking on the green circular checkbox on the left, as seen on the screenshot below.
Starting from here, all gallery views in Studio will display prediction scores and will be ranked to help you understand what your model has learnt. The images are ranked from the ones that are the less representative of the concept you've selected to the ones that are the most representative of that concept (according to the current model). Hence, the first images are the one that the model version has a hard time understanding, and depending on the difficulty, it is relevant to add similar images to your training dataset.
Here is the kind of view that you will be able to get when clicking on a particular concept.
You will also be able to visualize predictions on the Image views.
For classification and tagging projects, you will get a score next to each concept corresponding to the confidence score of the neural network for the given concept.
For detection projects, you will be able to activate the prediction and visualize the bounding boxes proposed by the neural network. Each bounding box comes with a confidence score that is displayed in the navigation bar. To display those predictions, you need to click on the eye icon next to PREDICTIONS in the navigation bar.
In addition, you may want to filter the predictions by using their score. By default, only predictions above the optimal threshold are displayed. You can change that setting using the range control show below:
By changing this value, you will display more or less predicted boxes. This setting is defined concept-wise.