Giters

coolzhao / Geo-SAM

A QGIS plugin tool using Segment Anything Model (SAM) to accelerate segmenting or delineating landforms in geospatial raster images.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Geo-SAM and colab

aloboa opened this issue · comments

commented

Not having a local GPU, I wonder if the following approach is feasible/advised:

  1. Run the Geo-SAM python encoder step using colab (eg, modifying https://colab.research.google.com/github/coolzhao/Geo-SAM/blob/dev/GeoSAM-Image-Encoder/examples/geosam-image-encoder.ipynb
  2. RUn Geo-SAM Segmentation in QGIS (assuming no GPU is needed).

Also, can the encoder Co-pilot be run in colab?

commented

Hi @aloboa ,

Yes, the image encoding process is known to be time-consuming, and we recommend using a high-performance GPU to run the Geo-SAM Encoder. The GeoSAM-Image-Encoder (Python package) is a standalone Python package that does not require QGIS as a dependency. You can directly run the Geo-SAM Encoder in Google Colab. Here is an example notebook: https://colab.research.google.com/github/coolzhao/Geo-SAM/blob/main/GeoSAM-Image-Encoder/examples/geosam-image-encoder.ipynb.

For Geo-SAM Segmentation and Copilot, the performance requirements are not high, and regular CPUs can handle them. They must be run in QGIS for a user-friendly GUI.