Assignment 2
Points:
- Sidewalks Huggingface dataset access and Fiftyone visualization: 25 points
- Training a MaskRCNN Segmentor: 50 points
Mask-RCNNModel Finetuning for Remote Sensing Applications
In this project you will be asked to segment sidewalks from satellite imagery. The importance of remote sensing and its associated job market is growing rapidly. Remote sensing powers weather prediction, agriculture, urban planning, defense and many other applications.
See this tutorial and read the overview sections so that you understand the nature of geospatial data. Visualization of such data can be done with VSCode plugins such as Geo Data Viewer and TIFF or with the desktop application of QGIS.
Access the dataset
Create an account in HuggingFace Hub and access the images associated with the dataset: https://huggingface.co/datasets/aegean-ai/sidewalks. Note that there are smaller versions of this dataset that is compatible with the json MS-COCO format.
Ensure that the cell below is successfully executed. To do so you need to insert before it a cell that allows you to login to the HF Hub. HF Datasets are also accessible with git and git-lfs.
Build a Fiftyone dataset
Use this guide to visualize the dataset with its masks.
Train a Mask-RCNN model
Use this guide to train a Mask-RCNN model on the dataset.