The minespline project is a library for grouping potential detections of landmines into curves. This library can also be used for grouping detections of any field of points.
The following example geojson files can be found in the data directory
MinelineHardCases1_1 geojson file
python -m minespline.mineliner evaluate MinelineHardCases1_1.geojson| Input | Output |
|---|---|
 |
 |
VeryHardWithoutDistractors_1 geojson file
python -m minespline.mineliner evaluate VeryHardWithoutDistractors_1.geojson| Input | Output |
|---|---|
 |
 |
VeryHardWithDistractors_4 geojson file
python -m minespline.mineliner evaluate VeryHardWithDistractors_4.geojson| Input | Output |
|---|---|
 |
 |
- Anaconda3, for python 3.7+ must be installed.
-
Obtain the sourcecode. If you have access to the git repo, then you can use
git clone https://github.com/HubbardJacob/MineSplines.git
-
Use anaconda to install the project dependancies. This will create our own virtual python environment called
ML(for Mine Lines).cd minespline conda env create- If you already have an
MLenvironment for this project, you should instead update it using
conda env update
- If you already have an
Before doing anything, make sure the ML environment is active
conda activate MLIf you want to run one of the command below from a script, you can use
conda run -n ML <Rest of the command goes here>which will run the script in the appropriate environment.
conda activate ML
# The most recent "very hard" examples
python -m minespline.mineliner evaluate ./data/test/very-hard/*.geojson
# Plot an example at a different threshold and throw away the output
python -m minespline.mineliner --max-spacing 9 --top-k 10 --threshold 0.00001 show data/test/RealCurvilinearData/RealDat_Curvi_1_1.geojson --plot -o /dev/null
From the command line interface (CLU)
(ML)$ python -m minespline.mineliner --help
Usage: mineliner.py [OPTIONS] COMMAND [ARGS]...
Options:
--min-length INTEGER The minimum length of a path [env var: ML_MIN_LENGTH;
default: 3]
--top-k INTEGER The number of paths to return (in descending order)
[env var: ML_TOP_K; default: 3]
--max-it INTEGER The max length of a path [env var: ML_MAX_IT;
default: 9223372036854775807]
--threshold FLOAT The minimum probability of a path [env var:
ML_THRESHOLD; default: 1e-35]
--spacing FLOAT The mine spacing, in meters. [env var: ML_SPACING;
default: 8]
--std-along FLOAT The standard deviation in mine spacing along the mine
line, in meters. [env var: ML_STD_ALONG; default:
0.2]
--std-across FLOAT The STD in mine spacing across the mine line, in
meters. [env var: ML_STD_ACROSS; default: 0.0625]
--max-spacing FLOAT The maximum distance between two mines [env var:
ML_MAX_SPACING; default: 17]
--prob-miss FLOAT The probability of a false-negative in the input [env
var: ML_PROB_MISS; default: 0.05]
--no-guess-spacing Do not guess the spacing -- instead use the value
passed in or specified in the file [env var:
ML_NO_GUESS_SPACING; default: False]
--no-progress Disable progress bars [env var: ML_NO_PROGRESS;
default: False]
--help Show this message and exit.
Commands:
evaluate Process many files and summarize results, saving a CSV and
plots...
show Process a single input file and save the results, optionally...
Will provide help that is most up to date. Most of the options allow you to set parameters that control mine processing.
At the time of this writing the CLI has two main commands:
(ML)$ python -m minespline.mineliner show --help
Usage: mineliner.py show [OPTIONS] INPUT_FILE
Process a single input file and save the results, optionally showing a
plot.
Options:
-o, --output FILENAME GeoJSON formatted output [env var: ML_SHOW_OUTPUT]
--plot Display a plot of the results [env var:
ML_SHOW_PLOT]
--help Show this message and exit.
(ML) $ python -m minespline.mineliner evaluate --help
Usage: mineliner.py evaluate [OPTIONS] [FILES]...
Process many files and summarize results, saving a CSV and plots of each
file.
Options:
-o, --outdir DIRECTORY Directory to save per-file results [env var:
ML_EVALUATE_OUTDIR]
--p-sep FLOAT Set the difference in means (-1 means leave alone)
[env var: ML_EVALUATE_P_SEP; default: -1]
--p-std FLOAT Set the std. of each class (before clipping to 0/1)
(-1 means leave alone) [env var:
ML_EVALUATE_P_STD; default: -1]
--drop FLOAT Fraction of positive examples to drop [env var:
ML_EVALUATE_DROP; default: 0]
--rep INTEGER Number of times to process each file [env var:
ML_EVALUATE_REP; default: 1]
--help Show this message and exit.
We operate on GEOJSON files and GEOTIFF files. Other formats used e.g. by COBRA are converted.
python -m minespline.ingest_cobra --helpGiven inital points with confidence values that can be interpreted as probabilities, we randomly sample a sequence of mines that form a markov chain. We assign a probability to the entire chain. The output is a single GEOJSON file with a linestring for the entire path, point features for each mine, and line features for each edge.
python -m minespline.sample --help