FishNet: AI For Fish Stock Estimation

The attached model was trained on a 63000 dataset of fish images belonging to 163 species. First, we trained a detectron2 model to detect and segment fish and fiduciary markers on a board. The detectron2 model was written in PyTorch, and the final classifier is a ResNet50 keras model.

Below is an example of how to use the 2 models.

Packages

Load Models

load an image and transform

run the segmentation model

visualize

classify the fish