For ocean environment protection, the government invested substantial time and effort in environmental protection, such as funding environment research and introducing new laws and regulations. Nowadays, artificial intelligence is considered an exact method because it may bring reforms to this area. Also, AI has acquired significant achievements, such as computer vision and deep learning fields. However, there are few publicly available, large scale underwater benchmark datasets. In this task, we first create such a dataset using videos captured by the SmartBay Ocean Observatory in Galway. Besides, a state-of-the-art deep model (Cascade Mask R-CNN) is re-trained using the benchmark dataset. The model achieved 72% mean average precision (mAP) on object detection (bounding box) and 66% mAP on the instance segmentation. Results show that the dataset created can provide sufficient information to train a deep model. Also, state-of-the-art deep models can detect and segment marine life with relatively high accuracy.
