Package containing helper functions for loading and evaluating DAVIS.
A Matlab version of the same package is also available.
DAVIS is released under the BSD License [see LICENSE for details]
DAVIS (Densely Annotated VIdeo Segmentation), consists of high quality, Full HD video sequences, spanning multiple occurrences of common video object segmentation challenges such as occlusions, motion-blur and appearance changes. Each video is accompanied by densely annotated, pixel-accurate and per-frame ground truth segmentation.
Execute the script ROOT/python/tools/eval.py
providing the resulting segmentation and setting the correct phase (train,val etc...) and year (2016,2017). See script documentation for mode details.
Example: python tools/eval.py -i path-to-my-technique -o results.yaml --year 2017 --phase val
Execute the script ROOT/python/tools/visualize.py
. The command-line arguments are similar to the evaluation script. Use --single-object
to visualize the original DAVIS 2016.
Example: python tools/visualize.py -i path-to-my-technique --year 2017 --phase val
C++
Python
- See ROOT/python/requirements.txt (Optionally to visualize results install
cv2
)
C++
- ./configure.sh && make -C build/release
Python:
- pip install virtualenv virtualenvwrapper
- source /usr/local/bin/virtualenvwrapper.sh
- mkvirtualenv davis
- pip install -r python/requirements.txt
- export PYTHONPATH=$(pwd)/python/lib
- See ROOT/python/lib/davis/config.py for a list of available options
See source code for documentation.
The directory is structured as follows:
-
ROOT/cpp
: Implementation and python wrapper of the temporal stability measure. -
ROOT/python/tools
: contains scripts for evaluating segmentation.eval.py
: evaluate a technique and store results in HDF5 fileeval_view.py
: read and display evaluation from HDF5.visulize.py
: visualize segmentation results.
-
ROOT/python/lib/davis
: library package contains helper functions for parsing and evaluating DAVIS -
ROOT/data
:get_davis.sh
: download input images and annotations.
Please cite DAVIS
in your publications if it helps your research:
@inproceedings{Perazzi_CVPR_2016,
author = {Federico Perazzi and
Jordi Pont-Tuset and
Brian McWilliams and
Luc Van Gool and
Markus Gross and
Alexander Sorkine-Hornung},
title = {A Benchmark Dataset and Evaluation Methodology for Video Object Segmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2016}
}
@article{Pont-Tuset_arXiv_2017,
author = {Jordi Pont-Tuset and
Federico Perazzi and
Sergi Caelles and
Pablo Arbel\'aez and
Alexander Sorkine-Hornung and
Luc {Van Gool}},
title = {The 2017 DAVIS Challenge on Video Object Segmentation},
journal = {arXiv:1704.00675},
year = {2017}
}
- Temporal stability measure (T)
- Per-attribute evaluation script
- Add usage examples