Skip to content

README.md

Latest commit

 

History

History
61 lines (42 loc) · 2.65 KB

README.md

File metadata and controls

61 lines (42 loc) · 2.65 KB

MISSFish

The Marine Institute and SmartBay underSea Fish dataset

Dataset:

Due to the limited space on Github, the annotated subset of videos (45 videos, ~3G) is hosted on Google Drive link.

Annotations are in data/annotations folder, currently the annotation is in MaskRCNN format. We are working on other formats, e.g. COCO.

TODO:

  • add a demo to show the annotation
  • add other standard annotation format such as COCO format
  • add sample images here

Live video feed can be accessed at: https://smartbay.marine.ie/

Historical videos (over 350k 2min video clips), can be accessed at here or click the download arrow on the right top corner of the live video feed.

Method

The Mask R-CNN used in this work is based on matterport repo link

A sample code for training a Mask R-CNN is available under the sample folder.

Trained Model

Currently, two trained MaskRCNN models are avaliable

  • retrained (training loss: 0.7314, validation loss:1.3885, validation mAP: 0.6228472, test mAP: 0.60101705): download
  • trained from scrach (training loss: 1.8018, validation loss: 1.6753, validation mAP: 0.09471174753137171, test mAP: 0.09028191691404833): download

Docker

To simplfy the setup, we used tensorflow/tensorflow:1.15.0rc2-gpu-py3-jupyter docker conainer.

Tools

Some tools, e.g. extract frames from video, are available.

Sample

TODO

  • add a Mask R-CNN training sample
  • add a demo to how to use existing model
  • add sample results here
  • add sample output video here

Funding

This research work is funded by the National Infrastructure Access Programme, which is funded by the Marine Institute under the Marine Research Programme with the support of the Irish Government. This publication has emanated from research supported in part by a research grant from Science Foundation Ireland (SFI) under Grant Number SFI/12/RC/2289_P2.

Citation

If you use this dataset in your research, please cite this project.

@article{missfish,
  title   = {Please holder},
  author  = {please holder},
  journal= {please holder},
  year={please holder}
}