Identify a Photo
Upload your image below and the system will attempt to identify it to the group that it belongs to.
Information about this feature:
- Images submitted are immediately discarded by the system. They are not retained for any purpose.
- All images that were referenced to train this feature are freely available and have been used in accordance with their creative commons license.
- This page should work for any marine organism worldwide, as long as it's from a group that features on this site, but you'll likely need to refer to another resource to drill down further.
- Any prediction with a confidence level over 10% will be shown in the results. If the highest confidence prediction is less than 10%, an empty set of results will be shown.
- The prediction model favours live, in-situ organisms. Wash-ups, especially if they are long dead, may not be identified as accurately.
- It's not trained for negative responses, so that means that any photo will be identified as a marine organism, even if it's not one (Fun game - does your dog look more like a seal or an acorn worm?).
- Certain fish families don't link correctly, and even though you are given correct information, the link will take you to the generic fishes page. This is due to the way I initially chose to handle fish families that are only represented by a single species on the site (ie, the butterflyfishes - Chaetodontidae).
- Any file under 4MB will be accepted by the system, but only bmp, jpg or png image files will be processed. Any other file type will result in an empty set of results.
- The AI model will improve over time, but its results won't always be accurate. Photo quality, similarity of organisms and other factors may result in inaccurate classification.
Current Vision Model Statistics
As of: 12/12/2024
Precision: If a prediction (of over 50% confidence) is made, how likely is it to be correct?
Recall: What is the likelihood of the correct prediction being made (at over 50% confidence)?
Note: This is based on the training data, and your results may vary
Other stats
Number of classifications (tags): 179
Target number of training images (per tag): 150
Number of classifications below image count target: 14 (7.8%)