Computer Vision
Test whether a model recognizes an object when its outline, proportions, or level of detail changes.
DATA TO LOOK FORLabeled objects and consistent representations
Compare shapes, predict categories, or match an image to a text label. These are starting points for experiments, not benchmark claims.
Test whether a model recognizes an object when its outline, proportions, or level of detail changes.
DATA TO LOOK FORLabeled objects and consistent representations
Train a classifier to sort icons into categories such as transport, technology, or food. Check where it confuses similar shapes.
DATA TO LOOK FORSemantic icon families and category labels
Compare how a model identifies the same object across different icon styles, from a simple outline to a geometric drawing.
DATA TO LOOK FORTechnology and transportation symbols
Use one icon as a query and retrieve similar objects. Compare matches based on appearance with matches based on meaning.
DATA TO LOOK FORTagged symbols and visual neighborhoods
Pair icons with labels or short descriptions, then test whether the model matches each image to the right text.
DATA TO LOOK FORCaptioned icons and descriptive metadata
Study whether generated icons follow a requested subject and drawing style. Compare the outputs with labeled examples.
DATA TO LOOK FORGeometric objects and style-controlled assets
Reserve a set of icons for evaluation. Compare results across categories and keep related variants out of both training and test sets.
DATA TO LOOK FORVersioned collections and evaluation subsets
Start with the subjects and your intended use. Add file formats and sample quantities when you have them.