PROCESSING AQUATIC LIFE IMAGES

1 . A computer-implemented method comprising:

receiving aquatic life data comprising a plurality of aquatic life images from a user through a user interface;

receiving, within the user interface, a first user request to use the aquatic life data to train a machine learning model configured to process an aquatic life image to classify the aquatic life image into one of multiple aquatic life categories;

determining a data curator score for each aquatic life image based on applying one or more of an object detection, clustering, or auto labeling processing to the plurality of aquatic life images;

identifying, based on the data curator scores, a proper subset of the plurality of aquatic life images;

providing the proper subset of the plurality of aquatic life images to one or more data annotators;

receiving annotation data generated by the one or more data annotators for the proper subset of the plurality of aquatic life images; and

providing the annotation data to a training system configured to train the machine learning model by using the annotation data and the proper subset of the plurality of aquatic life images to update parameter values of the machine learning model.

2 . The method of claim 1 , wherein the data curator score for each aquatic life image is an uncertainty score that is dependent on a likelihood that the aquatic life image will be misclassified by the machine learning model.

3 . The method of claim 2 , wherein identifying the proper subset of the plurality of aquatic life images comprises selecting aquatic life images that have uncertainty scores that satisfy an uncertainty score threshold.

4 . The method of claim 1 , wherein the data curator score for each aquatic life image is a rarity score that represents a number of aquatic life images having a same aquatic life category as the aquatic life image relative to a total number of the plurality of aquatic life images.

5 . The method of claim 4 , wherein identifying the proper subset of the plurality of aquatic life images comprises selecting aquatic life images that have rarity scores that satisfy a rarity score threshold.

6 . The method of claim 1 , further comprising:

receiving, within the user interface, a second user request to identify aquatic life images similar to a target aquatic life image;

determining a similarity score for each aquatic life image that represents a similarity of the aquatic life image relative to the target aquatic life image;

identifying one or more aquatic life images that satisfy a similarity score threshold; and

providing, through the user interface, the one or more aquatic life images that satisfy a similarity score threshold to the user.

7 . The method of claim 1 , further comprising:

receiving, within the user interface, a third user request to prioritize the plurality of aquatic life images for a downstream task;

determining a relevance score for each aquatic life image that represents a relevance of the aquatic life image relative to the downstream task;

generating a ranking of the plurality of aquatic life images based on the relevance scores; and

presenting, within the user interface, the ranking of the plurality of aquatic life images.

8 . The method of claim 7 , wherein the downstream task comprises one of: a fish disease detection task, a species identification task, a net inspection task, or a fish tracking task.

9 . A system comprising: one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:

receiving aquatic life data comprising a plurality of aquatic life images from a user through a user interface;

receiving, within the user interface, a first user request to use the aquatic life data to train a machine learning model configured to process an aquatic life image to classify the aquatic life image into one of multiple aquatic life categories;

determining a data curator score for each aquatic life image based on applying one or more of an object detection, clustering, or auto labeling processing to the plurality of aquatic life images;

identifying, based on the data curator scores, a proper subset of the plurality of aquatic life images;

providing the proper subset of the plurality of aquatic life images to one or more data annotators;

receiving annotation data generated by the one or more data annotators for the proper subset of the plurality of aquatic life images; and

providing the annotation data to a training system configured to train the machine learning model by using the annotation data and the proper subset of the plurality of aquatic life images to update parameter values of the machine learning model.

10 . The system of claim 9 , wherein the data curator score for each aquatic life image is an uncertainty score that is dependent on a likelihood that the aquatic life image will be misclassified by the machine learning model.

11 . The system of claim 10 , wherein identifying the proper subset of the plurality of aquatic life images comprises selecting aquatic life images that have uncertainty scores that satisfy an uncertainty score threshold.

12 . The system of claim 9 , wherein the data curator score for each aquatic life image is a rarity score that represents a number of aquatic life images having a same aquatic life category as the aquatic life image relative to a total number of the plurality of aquatic life images.

13 . The system of claim 12 , wherein identifying the proper subset of the plurality of aquatic life images comprises selecting aquatic life images that have rarity scores that satisfy a rarity score threshold.

14 . The system of claim 9 , wherein the operations further comprise:

receiving, within the user interface, a second user request to identify aquatic life images similar to a target aquatic life image;

determining a similarity score for each aquatic life image that represents a similarity of the aquatic life image relative to the target aquatic life image;

identifying one or more aquatic life images that satisfy a similarity score threshold; and

providing, through the user interface, the one or more aquatic life images that satisfy a similarity score threshold to the user.

15 . The system of claim 9 , wherein the operations further comprise:

receiving, within the user interface, a third user request to prioritize the plurality of aquatic life images for a downstream task;

determining a relevance score for each aquatic life image that represents a relevance of the aquatic life image relative to the downstream task;

generating a ranking of the plurality of aquatic life images based on the relevance scores; and

presenting, within the user interface, the ranking of the plurality of aquatic life images.

16 . The system of claim 15 , wherein the downstream task comprises one of: a fish disease detection task, a species identification task, a net inspection task, or a fish tracking task.

17 . A computer storage medium encoded with a computer program, the program comprising instructions that are operable, when executed by data processing apparatus, to cause the data processing apparatus to perform operations comprising:

receiving aquatic life data comprising a plurality of aquatic life images from a user through a user interface;

receiving, within the user interface, a first user request to use the aquatic life data to train a machine learning model configured to process an aquatic life image to classify the aquatic life image into one of multiple aquatic life categories;

determining a data curator score for each aquatic life image based on applying one or more of an object detection, clustering, or auto labeling processing to the plurality of aquatic life images;

identifying, based on the data curator scores, a proper subset of the plurality of aquatic life images;

providing the proper subset of the plurality of aquatic life images to one or more data annotators;

receiving annotation data generated by the one or more data annotators for the proper subset of the plurality of aquatic life images; and

providing the annotation data to a training system configured to train the machine learning model by using the annotation data and the proper subset of the plurality of aquatic life images to update parameter values of the machine learning model.

18 . The computer storage medium of claim 17 , wherein the operations further comprise:

receiving, within the user interface, a second user request to identify aquatic life images similar to a target aquatic life image;

determining a similarity score for each aquatic life image that represents a similarity of the aquatic life image relative to the target aquatic life image;

identifying one or more aquatic life images that satisfy a similarity score threshold; and

providing, through the user interface, the one or more aquatic life images that satisfy a similarity score threshold to the user.

19 . The computer storage medium of claim 17 , wherein the operations further comprise:

receiving, within the user interface, a third user request to prioritize the plurality of aquatic life images for a downstream task;

determining a relevance score for each aquatic life image that represents a relevance of the aquatic life image relative to the downstream task;

generating a ranking of the plurality of aquatic life images based on the relevance scores; and

presenting, within the user interface, the ranking of the plurality of aquatic life images.

20 . The computer storage medium of claim 19 , wherein the downstream task comprises one of: a fish disease detection task, a species identification task, a net inspection task, or a fish tracking task.