Butterfly Detector

Butterfly Detector for Aerial Images

Current state-of-the-art object detectors have achieved high performance when applied to images captured by standard front facing cameras. When applied to high-resolution aerial images captured from a drone or UAV stand-point, they fail to generalize to the wide range of objects' scales. In order to address this limitation, we propose an object detection method called Butterfly Detector that is tailored to detect objects in aerial images. We extend the concept of fields and introduce butterfly fields, a type of composite field that describes the spatial information of output features as well as the scale of the detected object. To overcome occlusion and viewing angle variations that can hinder the localization process, we employ a voting mechanism between related butterfly vectors pointing to the object center. We evaluate our Butterfly Detector on two publicly available UAV datasets (UAVDT and VisDrone2019) and show that it outperforms previous state-of-the-art methods while remaining real-time.

Demo

Setup

Python 3 is required. Python 2 is not supported. Do not clone this repository and make sure there is no folder named butterflydetector in your current directory.

pip3 install butterflydetector

For development of the butterflydetector source code itself, you need to clone this repository and then:

pip3 install numpy cython
pip3 install --editable '.[train,test]'

The last command installs the Python package in the current directory (signified by the dot) with the optional dependencies needed for training and testing.

Data structure

data         
├── UAV-benchmark-M
    ├── test
    ├── train
├── VisDrone2019
    ├── VisDrone2019-DET-train
        ├── annotations
        ├── images
    ├── VisDrone2019-DET-val
    ├── VisDrone2019-DET-test-dev

Interfaces

• python3 -m butterflydetector.predict --help
• python3 -m butterflydetector.train --help
• python3 -m butterflydetector.eval --help
• python3 -m butterflydetector.logs --help

Tools to work with models:

• python3 -m butterflydetector.migrate --help

Benchmark

Comparison of AP (%), False Positives (FP), and Recall (%) with state-of-the-art methods on UAVDT datasets.

Comparison of AP (average precision), AR (average recall), True Posi-tives (TP), and False Positives (FP) with state-of-the-art methods on VisDronedataset

Visualization

To visualize logs:

python3 -m butterflydetector.logs \
  outputs/<model1-basename>.pkl.log \
  outputs/<model2-basename>.pkl.log \
  outputs/<model3-basename>.pkl.log

Train

See datasets for setup instructions.

The exact training command that was used for a model is in the first line of the training log file.

Train a HRNetW32-det model on the VisDrone Dataset:

time CUDA_VISIBLE_DEVICES=0,1 python3 -m butterflydetector.train \
  --lr=1e-3 \
  --momentum=0.95 \
  --epochs=150 \
  --lr-decay 120 140 \
  --batch-size=5 \
  --basenet=hrnetw32det \
  --headnets butterfly10 \
  --square-edge=512 \
  --lambdas 1 1 1 1 \
  --dataset visdrone \
  --butterfly-side-length -2

You can refine an existing model with the --checkpoint option.

Evaluation

The command below will run your model on visdrone and save the predictions in the output directory. The predictions are saved in the correct format to be read by the official Matlab evaluator of VisDrone2019. To evaluate on UAVDT, simply replace 'visdrone' to 'uavdt'.

python -m butterflydetector.eval --checkpoint <directory-to-checkpoint> --dataset visdrone --output <directory-to-store-predictions> --seed-threshold 0.1

Video

Processing a video frame by frame from video.avi to video.pose.mp4 using ffmpeg:

export VIDEO=video.avi  # change to your video file

mkdir ${VIDEO}.images
ffmpeg -i ${VIDEO} -qscale:v 2 -vf scale=641:-1 -f image2 ${VIDEO}.images/%05d.jpg
python3 -m butterflydetector.predict --checkpoint resnet152 --glob "${VIDEO}.images/*.jpg"
ffmpeg -framerate 24 -pattern_type glob -i ${VIDEO}.images/'*.jpg.skeleton.png' -vf scale=640:-2 -c:v libx264 -pix_fmt yuv420p ${VIDEO}.pose.mp4

In this process, ffmpeg scales the video to 641px which can be adjusted.

EPFL Roundabout Dataset

EPFL Roundabout is a dataset that contains more than 2 hours of drone data collected from 4 different roundabout locations: Morges, EPFL, Ecublens, Echandens. The dataset contains both detection and tracking labels for 6 different categories: car, truck, bus, van, cyclists, pedestrians. The dataset can be used for detection, tracking, as well as trajectory prediction. The latter is usually challenging at a roundabout. The dataset can be downloaded here

Citation

@misc{adaimi2020perceiving,
      title={Perceiving Traffic from Aerial Images},
      author={George Adaimi and Sven Kreiss and Alexandre Alahi},
      year={2020},
      eprint={2009.07611},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}