PolyDETR: Polyline Detection Transformer

PolyDETR's goal is to detect polylines' location in the topographic maps

Docker imagery to train/testing PolyDETR

Here is the command to run the docker imagery

sudo nvidia-docker run -t -i -v {local_dir}:{docker_dir} -p 8888:8888 pytorch/pytorch:1.2-cuda10.0-cudnn7-devel

Input topographic maps

The topographic maps from the map competition. The competition provides tif map images, and tif label images. Label images are binary {0, 1}, 1 reprensents the pixels belong to the desired polyline feature.

Train PolyDETR

The training process includes

1. data processing: convert tif map images to png images, and tif label image to vector (shapefile). The png map images and vector polylines are training data
2. data augmentation: includes shifting along x- and y-axis, and rotation every 90 degrees.
3. training PolyDETR

To update the parameters for data processing, model architecture, and training process, please update './util/args.py'

**Here is the command to train PolyDETR python train.py

Use PolyDETR to detect desired polylines

Update './util/args_test.py' to set the map name and trained model path for PolyDETR

Here is the command to test PolyDETR

python test.py

Here is the command to use PolyDETR to detect line features for the competition

python test_main_competition.py Please update './util/args_test_multimaps.py' to set directory for the map images