This tool helps us collect data on whether photos and sketches match to train better deep learning models for multi-modal image understanding.
Your Task
For each pair of images shown, you need to decide if the photo and sketch represent the same subject. You can determine the match by selecting one of the following options:
Matching Pairs (✓)
Photo and sketch clearly represent the same subject. Usually the dominant objects are similar in shape, size, and position. Sometimes there are not as many objects in the sketch (e.g., if the sketch is a simplified version of the photo).
Uncertain Pairs (?)
Use "Maybe" when the relationship is ambiguous. The orientation or scale of the objects may differ, the count is much different or the background is unrecognizable.
Non-Matching Pairs (✗)
Photo and sketch represent different subjects. Typically, the objects in the photo and sketch are clearly not the same.
After selecting an option, a new pair will load automatically. Your responses are saved immediately.
Tips
- Take a few moments to examine both images carefully, sometimes different proportions or perspectives can make it harder to tell if they match or not.
- Use "Maybe" when you're not confident about your answer, you can use this option frequently when unsure rather than being strict.
- Try to be consistent in your judgments across different pairs to help us gather reliable data. Note that the system is callibrated in a way that about half of samples will be at least somewhat semantically similar.