PyCFTrackers
Python re-implementation of some correlation filter based tracker
Install / Use
/learn @fengyang95/PyCFTrackersREADME
pyCFTrackers
Python re-implementation of some correlation filter based tracker, and all of these algorithms are implemented based on the official Matlab code. All the code has been tested on Ubuntu 16.04, Python 3.5. I use pysot-toolkit to eval the performance on OTB and VOT.
- [x] MOSSE
- [x] CSK
- [x] CN
- [x] KCF/DCF
- [x] DSST/DSST-LP
- [x] SAMF
- [x] Staple
- [x] Staple-CA
- [x] BACF
- [x] CSR-DCF/CSR-DCF-LP
- [x] DAT This tracker is not based on CF, I re-implement it just for understanding Staple better
- [x] LDES
- [x] MKCFup/MKCFup-LP
- [x] MCCTH
- [x] STRCF
Trackers from other repos - [x] ECO/ECO-HC from pyECO
- [x] OPENCV-CSRDCF(Note that trackers implemented in opencv get lower score because they will return a success flag and if this flag is False, the bbox will be set to (0,0,0,0))
Install
git clone https://github.com/wwdguu/pyCFTrackers.git && cd pyCFTrackers
export pyCFTrackers=$PWD
pip install -r requirements.txt
cd lib/eco/features/
python setup.py build_ext --inplace
cd ../../..
cd lib/pysot/utils/
python setup.py build_ext --inplace
cd ../../..
export PYTHONPATH=$PWD:$PYTHONPATH
Get Dataset
You can follow the instructions in the following repo to get VOT2016,VOT2018 and OTB100 dataset.
trackdat
Then get the json files according to pysot-toolkit
Then put the data in the dataset dir.
Demo
cd examples
python cf_demo.py
Eval on OTB and VOT
cd eval
python get_vot2016_result.py
python get_vot2018_result.py
python ope_otb.py
python eval_VOT2016.py
python eval_VOT2018.py
python eval_OTB.py
OTB result
OTB-100
| OTB100 Success Plot | OTB100 Precision Plot |
| --------------------------------- | ----------------------------- |
|
|
|
OTB-2013
| OTB2013 Success Plot | OTB2013 Precision Plot |
| --------------------------------- | ----------------------------- |
|
|
|
VOT result
VOT2018
---------------------------------------------------------------
| Tracker Name | Accuracy | Robustness | Lost Number | EAO |
---------------------------------------------------------------
| ECO | 0.485 | 0.403 | 86.0 | 0.224 |
| CSRDCF | 0.492 | 0.501 | 107.0 | 0.210 |
| ECO-HC | 0.500 | 0.473 | 101.0 | 0.207 |
| CSRDCF-LP | 0.503 | 0.553 | 118.0 | 0.199 |
| Staple | 0.524 | 0.665 | 142.0 | 0.179 |
| LDES | 0.528 | 0.684 | 146.0 | 0.175 |
| MCCTH-Staple | 0.535 | 0.684 | 146.0 | 0.172 |
| OPENCV-CSRDCF | 0.486 | 0.651 | 139.0 | 0.170 |
| BACF | 0.511 | 0.674 | 144.0 | 0.169 |
| DAT | 0.477 | 0.777 | 166.0 | 0.158 |
| STRCF | 0.483 | 0.679 | 145.0 | 0.152 |
| CN | 0.439 | 1.100 | 235.0 | 0.112 |
| SAMF | 0.499 | 1.147 | 245.0 | 0.110 |
| DSST | 0.492 | 1.222 | 261.0 | 0.107 |
| DSST-LP | 0.512 | 1.260 | 269.0 | 0.103 |
| DCF | 0.463 | 1.246 | 266.0 | 0.099 |
| KCF | 0.463 | 1.330 | 284.0 | 0.094 |
| CSK | 0.418 | 1.386 | 296.0 | 0.090 |
| MOSSE | 0.378 | 1.967 | 420.0 | 0.063 |
---------------------------------------------------------------
VOT2016
---------------------------------------------------------------
| Tracker Name | Accuracy | Robustness | Lost Number | EAO |
---------------------------------------------------------------
| ECO | 0.564 | 0.256 | 55.0 | 0.336 |
| MCCTH-Staple | 0.574 | 0.359 | 77.0 | 0.303 |
| Staple | 0.560 | 0.387 | 83.0 | 0.299 |
| ECO-HC | 0.532 | 0.350 | 75.0 | 0.293 |
| CSRDCF | 0.542 | 0.359 | 77.0 | 0.273 |
| LDES | 0.577 | 0.419 | 90.0 | 0.272 |
| CSRDCF-LP | 0.548 | 0.354 | 76.0 | 0.272 |
| BACF | 0.521 | 0.405 | 87.0 | 0.252 |
| OPENCV-CSRDCF | 0.521 | 0.438 | 94.0 | 0.239 |
| STRCF | 0.520 | 0.415 | 89.0 | 0.239 |
| DAT | 0.474 | 0.503 | 108.0 | 0.232 |
| SAMF | 0.544 | 0.639 | 137.0 | 0.193 |
| DSST-LP | 0.543 | 0.727 | 156.0 | 0.180 |
| CN | 0.468 | 0.653 | 140.0 | 0.178 |
| DSST | 0.531 | 0.732 | 157.0 | 0.177 |
| DCF | 0.474 | 0.704 | 151.0 | 0.171 |
| KCF | 0.469 | 0.718 | 154.0 | 0.171 |
| CSK | 0.433 | 0.886 | 190.0 | 0.139 |
| MOSSE | 0.388 | 1.244 | 267.0 | 0.096 |
---------------------------------------------------------------
License.
Licensed under an MIT license.
