SafeUAV: Learning to estimate depth and safe landing areas for UAVs from synthetic data

This repository holds the implementation of the paper, presented at the UAVision2018 workshop (ECCV).

https://sites.google.com/site/aerialimageunderstanding/safeuav-learning-to-estimate-depth-and-safe-landing-areas-for-uavs (for checkpoints, paper and other information).

Standard variables

model=unet_tiny_sun/unet_big_concatenate/deeplabv3plus/unet_classic (pick one)
dir=test_dir
lr=0.001
patience=4
factor=0.1
num_epochs=100

Training a model

HVO

python main.py train classification /path/to/dataset.h5 --model=$model --dir=$dir --label_dims=hvn_gt_p1 --batch_size=N --optimizer=Adam --learning_rate=$lr --patience=$patience --factor=$factor --num_epochs=$num_epochs

Depth

python main.py train regression /path/to/dataset.h5 --model=$model --dir=$dir --label_dims=depth --batch_size=N --optimizer=Adam --learning_rate=$lr --patience=$patience --factor=$factor --num_epochs=$num_epochs

Running a pre-trained model

HVO

python main.py test classification /path/to/dataset.h5 --model=$model --weights_file=/path/to/checkpoint.pkl --test_plot_results=1 --label_dims=hvn_gt_p1 --batch_size=N

Depth

python main.py test regression /path/to/dataset.h5 --model=$model --weights_file=/path/to/checkpoint.pkl --test_plot_results=1 --label_dims=depth --batch_size=N

Running on an existing video (outputs another video)

HVO

python main_inference_video.py classification in_video.mp4 out_video.mp4 --model=$model --weights_file=/path/to/checkpoint.pkl

Depth

python main_inference_video.py regression in_video.mp4 out_video.mp4 --model=$model --weights_file=/path/to/checkpoint.pkl

Installation for Jetson

$ sudo apt update && sudo apt -y upgrade
$ sudo apt -y install < packages.txt

$ pip3 install --upgrade pip
$ pip3 install -r requirements.txt

$ cd ..
$ git clone https://gitlab.com/mihaicristianpirvu/neural-wrappers.git
$ cd neural-wrappers
$ git checkout 3dcc404b08f0e356904d1a1dd16382c3ae4aa752

Use Docker Image

Docker version >= 20.10.

You can download pre-train model, datasets and Docker image here.

$ docker build -t <repository> docker/

Run

$ docker run --privileged --gpus all --it -v <data>:/SafeUAV/SafeUAV/data <repository>