This repo contains the code for the solution proposed by Gaussian Solutions LLC for the DARPA's challenge 'AI for Critical Mineral Assessment Competition'. More about the competition can be read [here] https://criticalminerals.darpa.mil/Leaderboard
Please refer the Technical brief for an overview of the approach.
conda create -n env_cma python=3.9
conda activate env_cma
git clone https://github.com/sureshgaussian/ai4cma.git
cd ai4cma
pip install -r requirements.txt
All the parameters required to process the inputs, run the experiments, generate predictions are defined in config.py file.
Specify the ROOT_PATH in config.py. This directory is used to hold the raw and preprocessed data, model checkpoints, and results.
Extract the training and validation data to $ROOT_PATH/data/
ROOT_PATH
├── data
│ ├── training
│ ├── validation
Run the below script to generate the input descriptors and tiled inputs.
python prepare.py -d challenge -o prepare_inputs
This creates a folder structure as shown below
ROOT_PATH
├── tiled_inputs
│ ├── info
│ │ ├── challenge_training_files.csv (holds meta data of each binary raster file)
│ │ ├── challenge_testing_files.csv
│ ├── challenge_training
│ │ ├── info
│ │ │ ├── all_tiles.csv (holds the tile information)
│ │ │ ├── balanced_tiles.csv
│ │ ├── inputs
│ │ ├── legends
│ │ ├── masks
│ ├── challenge_testing
│ │ ├── info
│ │ │ ├── all_tiles.csv (holds the tile information)
│ │ │ ├── balanced_tiles.csv
│ │ ├── inputs
│ │ ├── legends
│ │ ├── masks
│ ├── ...
Parameters related to training are specified in $ROOT_PATH/config.py.
python train.py -d challenge
Saves the model checkpoints under $ROOT_PATH/temp.
Download the trained models from here and place them under $ROOT_PATH/temp.
Run the script to generate tiles for each validation file, run the inference and save the stitched predictions under $ROOT_PATH/results/
Note: For any inference task, please place the files under 'validation' folder for the script to pick up the files. This is a limitation of how we coded the path set up, relative paths etc, and not the algorithm limitation.
python inference.py -d challenge -s validation
ROOT_PATH
├── results
│ ├── testing
│ ├── validation
│ │ ├── ArthurTaylor_1990_Fig28_contour_line.tif
│ │ ├── pp1410b_Lower_Mesozoic_diaba_pt.tif
│ │ .
│ │ .
Step-1 of post processing to remove false positives within the map region. Erodes the line predictions
python postprocessing.py
ROOT_PATH
├── results_pp_within_map
│ ├── testing
│ ├── validation
│ │ ├── ArthurTaylor_1990_Fig28_contour_line.tif
│ │ ├── pp1410b_Lower_Mesozoic_diaba_pt.tif
│ │ .
│ │ .
Step-2 of post-processing to remove false positives outside the map region.
python filter_non_map_region/inference_map.py
python postprocessing_submission.py
ROOT_PATH
├── results_pp_outside_map
│ ├── testing
│ ├── validation
│ │ ├── ArthurTaylor_1990_Fig28_contour_line.tif
│ │ ├── pp1410b_Lower_Mesozoic_diaba_pt.tif
│ │ .
│ │ .
Run the below command for qualitative analysis of predictions. Sample visualizations are available in $ROOT_PATH/sample_viz.
python generate_visualizations.py
ROOT_PATH
├── results_contours
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent