The code in this repository was used to generate the results of a benchmark of general purpose tracking algorithms on the maritime setting using airborne imagery. We used the OTB framework [1].
We also present a new approach [2] which is based on KCF [3] tracker and blob analysis. The evaluations are done either with CNN [4] or HOG [5] features.
Requirements for the evaluation of all methods.
- Matlab (2015a used)
- Python 2.7
- Numpy
- OpenCV 3.1
- Caffe
- Dlib 18.18
- VLFeat (for OTB)
- Matconvnet (for CF2 and MDNet)
- Mexopencv (for MUSTer)
Airborne Maritime Dataset.
Our method:
- Setup Python 2.7 with Numpy
- Install OpenCV 3.1 (Tutorial)
- Install Caffe with CUDA (Tutorial)
- Setup CAFFE_ROOT environment variable to the folder that contains the Caffe framework and models.
- Download VGG-Net [4] from https://gist.github.com/ksimonyan/3785162f95cd2d5fee77 and put it into the '.../caffe/model/' folder.
- Download the Maritime Dataset from https://www.dropbox.com/s/a737bzk7uktplu4/data_seq.zip?dl=0
- For OURS_HOG you might need to recompile the HOG extraction code. To do this edit the setup.py file to point to your dlib folder and then run in the terminal: python setup.py
Other algorithms:
Usually the other algorithms should run if the required libraries are correctly installed. Either way each tracker has a readme file in its folder.
The CF2 [6] tracker requires that you download the ConvNet model and put it into '/trackers/CF2/model/' https://uofi.box.com/shared/static/kxzjhbagd6ih1rf7mjyoxn2hy70hltpl.mat
If you have any questions: jorgep.s.matos@gmail.com
[1] Wu, Yi, Jongwoo Lim, and Ming-Hsuan Yang. "Online object tracking: A benchmark." Proceedings of the IEEE conference on computer vision and pattern recognition. 2013.
[2] J. Matos, A. Bernardino, and R. Ribeiro, “Robust tracking of vessels in oceanographic airborne images,” in OCEANS’16 MTS/IEEE Monterey. MTS/IEEE.
[3] Henriques, João F., et al. "High-speed tracking with kernelized correlation filters." IEEE Transactions on Pattern Analysis and Machine Intelligence 37.3 (2015): 583-596.
[4] Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014).
[5] Felzenszwalb, Pedro F., et al. "Object detection with discriminatively trained part-based models." IEEE transactions on pattern analysis and machine intelligence 32.9 (2010): 1627-1645.
[6] Ma, Chao, et al. "Hierarchical convolutional features for visual tracking." Proceedings of the IEEE International Conference on Computer Vision. 2015.