Illegal, Unreported and Unregulated (IUU) fishing is a threat to food supply and marine ecosystem health, It exacerbates the effects of climate change and can damage wildlife.
This demo is based on the xView3 challenge, in which scientists are being tasked to develop algorithms that identify and classify SAR satellite images of vessels suspected of engaging in IUU fishing.
Most illegal fishing vessels are not cooperating with automatic identification systems, so there must be some other method to detect them.
Today, GPUs can only handle images lower than 4k resolution.
In the below example, we include detections from models trained on image patches which are 800x800 (left), as well as larger patches which are 2048x2048 pixels(middle), all he way to 5k x 5k (right). Having more contextual information through higher resolution, is critical to making more accurate predictions.
Only Sambanova can handle true resolution data, which is critical to accuracy in object detection, segmentation problems, and image classification.