geo-inference is a Python package designed for feature extraction from geospatial imagery using compatible deep learning models. It provides a convenient way to extract features from large TIFF images and save the output mask as a TIFF file. It also supports converting the output mask to vector format (file_name.geojson), YOLO format (file_name.csv), and COCO format (file_name.json). This package is particularly useful for applications in remote sensing, environmental monitoring, and urban planning.
Geo-inference requires Python 3.11.
To install the package, use:
pip install geo-inference
The recipe to use cuda-enabled Geo-inference on Windows OS is slightly different than on Linux-based OS.
- Validate the nvidia drivers version installed on your computer by running
nvcc --version:
PS C:\> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:28:36_Pacific_Standard_Time_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0Note: If the command returns an error, you need to download and install the Nvidia-drivers first at https://developer.nvidia.com/cuda-downloads.
- Install pytorch-cuda following one method suggested here: https://pytorch.org/get-started/locally/.
Note: Make sure to select the cuda version matching the driver installed on your computer.
- Test the installation:
PS C:\> python
>>> import torch
>>> torch.cuda.is_available()
>>> True- install geo-inference using
pip:
pip install geo-inference
Alternatively, you can build the Dockerfile to use Geo-Inference.
Input: GeoTiffs with compatible TorchScript model. For example: A pytorch model trained on high resolution geospatial imagery with the following features:
- pixel size (0.1m to 3m)
- data type (uint8)
expects an input image with the same features. An example notebook for how the package is used is provided in this repo.
Here's an example of how to use Geo Inference (Command line and Script):
Command line
geo_inference -a <args>-a,--args: Path to arguments stored in yaml, consult ./config/sample_config.yaml
geo_inference -i <image> -br <bands_requested> -m <model> -wd <work_dir> -ps <patch_size> -v <vec> -d <device> -id <gpu_id> -cls <classes> -mg <mgpu> -pr <pr_thr>-i,--image: Path to Geotiff-bb,--bbox: AOI bbox in this format "minx, miny, maxx, maxy" (Optional)-br,--bands_requested: The requested bands from provided Geotiff (if not provided, it uses all bands)-m,--model: Path or URL to the model file-wd,--work_dir: Working Directory-ps,--patch_size: The patch Size, the size of dask chunks, Default = 1024-w,--workers: Number of workers used by dask, Default = Nb of cores available on the host, minus 1-v,--vec: Vector Conversion-y,--yolo: Yolo Conversion-c,--coco: Coco Conversion-d,--device: CPU or GPU Device-id,--gpu_id: GPU ID, Default = 0-cls,--classes: The number of classes that model outputs, Default = 5-mg,--mgpu: Whether to use multi-gpu processing or not, Default = False-pr,--prediction_thr: Prediction probability Threshold (fraction of 1) to use. Default = 0.3
You can also use the -h option to get a list of supported arguments:
geo_inference -hImport script
from geo_inference.geo_inference import GeoInference
# Initialize the GeoInference object
geo_inference = GeoInference(
model="/path/to/segformer_B5.pt",
work_dir="/path/to/work/dir",
mask_to_vec=False,
mask_to_yolo=False,
mask_to_coco=False,
device="gpu",
multi_gpu=False,
gpu_id=0,
num_classes=5,
prediction_threshold=0.3
)
# Perform feature extraction on a TIFF image
image_path = "/path/to/image.tif"
bands_requested = [1,2,3]
patch_size = 1024
workers = 0
patch_size = 512
bbox = "0, 0, 1000, 1000"
geo_inference(
inference_input = image_path,
bands_requested = bands_requested,
patch_size = patch_size,
workers = workers,
bbox=bbox
)Initiating the GeoInference class takes the following parameters:
model: The path or URL to the model file (.pt for PyTorch models) to use for feature extraction.work_dir: The path to the working directory. Default is"~/.cache".mask_to_vec: If set to"True", vector data will be created from mask. Default is"False"mask_to_yolo: If set to"True", vector data will be converted to YOLO format. Default is"False"mask_to_coco: If set to"True", vector data will be converted to COCO format. Default is"False"device: The device to use for feature extraction. Can be"cpu"or"gpu". Default is"gpu".multi_gpu: If set to"True", uses multi-gpu for running the inference. Default is"False"gpu_id: The ID of the GPU to use for feature extraction. Default is0.num_classes: The number of classes that the TorchScript model outputs. Default is5.prediction_threshold: Prediction probability Threshold (fraction of 1) to use. Default is0.3.
Calling the GeoInference object takes the following parameters:
inference_input: Path to Geotiff.bands_requested: The requested bands from provided Geotiff (if not provided, it uses all bands).patch_size: The patch size to use for feature extraction. Default is1024.workers: Number of workers used by Dask, Default is0= Number of cores available on the host, minus 1.bbox: AOI bbox in this format "minx, miny, maxx, maxy", in the image's crs. Default isNone.
The GeoInference class outputs the following files:
mask.tif: The output mask file in TIFF format.polygons.geojson: The output polygon file in GeoJSON format. This file is only generated if themask_to_vecparameter is set toTrue.yolo.csv: The output YOLO file in CSV format. This file is only generated if themask_to_vec,vec_to_yoloparameters are set toTrue.coco.json: The output COCO file in JSON format. This file is only generated if themask_to_vec,vec_to_cocoparameters are set toTrue.
Each file contains the extracted features from the input geospatial imagery.
Geo Inference is released under the Open Government License - Canada. See LICENSE for more information.
For any questions or concerns, please open an issue on GitHub.