pysot-toolkit

The purpose of this repo is to provide evaluation API of Current Single Object Tracking Dataset, including

• VOT2016

• VOT2018

• VOT2018-LT

• OTB100(OTB2015)

• UAV123

• NFS

• LaSOT

• TrackingNet (Evaluation on Server)

• GOT-10k (Evaluation on Server)

Install

git clone https://github.com/StrangerZhang/pysot-toolkit
pip install -r requirements.txtcd pysot/utils/
python setup.py build_ext --inplace# if you need to draw graph, you need latex installed on your system

Download Dataset

Download json files used in our toolkit baidu pan or Google Drive

1. Put CVRP13.json, OTB100.json, OTB50.json in OTB100 dataset directory (you need to copy Jogging to Jogging-1 and Jogging-2, and copy Skating2 to Skating2-1 and Skating2-2 or using softlink)

   The directory should have the below format

   | -- OTB100/

   | -- Basketball

   | ......

   | -- Woman

   | -- OTB100.json

   | -- OTB50.json

   | -- CVPR13.json

2. Put all other jsons in the dataset directory like in step 1

Usage

1. Evaluation on VOT2018(VOT2016)

cd /path/to/pysot-toolkit
python bin/eval.py 
	--dataset_dir /path/to/dataset/root 		# dataset path
	--dataset VOT2018 				# dataset name(VOT2018, VOT2016)
	--tracker_result_dir /path/to/tracker/dir 	# tracker dir
	--trackers ECO UPDT SiamRPNpp 			# tracker names # you will see------------------------------------------------------------|Tracker Name| Accuracy | Robustness | Lost Number |  EAO  |------------------------------------------------------------| SiamRPNpp  |  0.600   |   0.234    |    50.0     | 0.415 ||    UPDT    |  0.536   |   0.184    |    39.2     | 0.378 ||    ECO     |  0.484   |   0.276    |    59.0     | 0.280 |------------------------------------------------------------

2. Evaluation on OTB100(UAV123, NFS, LaSOT)

converted *.txt tracking results will be released soon

cd /path/to/pysot-toolkit
python bin/eval.py 
	--dataset_dir /path/to/dataset/root 		# dataset path
	--dataset OTB100 				# dataset name(OTB100, UAV123, NFS, LaSOT)
	--tracker_result_dir /path/to/tracker/dir 	# tracker dir
	--trackers SiamRPN++ C-COT DaSiamRPN ECO  	# tracker names 
	--num 4 				  	# evaluation thread
	--show_video_level  	  			# wether to show video results
	--vis 					  	# draw graph# you will see (Normalized Precision not used in OTB evaluation)-----------------------------------------------------|Tracker name| Success | Norm Precision | Precision |-----------------------------------------------------| SiamRPN++  |  0.696  |     0.000      |   0.914   ||    ECO     |  0.691  |     0.000      |   0.910   ||   C-COT    |  0.671  |     0.000      |   0.898   || DaSiamRPN  |  0.658  |     0.000      |   0.880   |-----------------------------------------------------

-----------------------------------------------------------------------------------------|    Tracker name     |      SiamRPN++      |      DaSiamRPN      |         ECO         |-----------------------------------------------------------------------------------------|     Video name      | success | precision | success | precision | success | precision |-----------------------------------------------------------------------------------------|     Basketball      |  0.423  |   0.555   |  0.677  |   0.865   |  0.653  |   0.800   ||        Biker        |  0.728  |   0.932   |  0.319  |   0.448   |  0.506  |   0.832   ||        Bird1        |  0.207  |   0.360   |  0.274  |   0.508   |  0.192  |   0.302   ||        Bird2        |  0.629  |   0.742   |  0.604  |   0.697   |  0.775  |   0.882   ||      BlurBody       |  0.823  |   0.879   |  0.759  |   0.767   |  0.713  |   0.894   ||      BlurCar1       |  0.803  |   0.917   |  0.837  |   0.895   |  0.851  |   0.934   ||      BlurCar2       |  0.864  |   0.926   |  0.794  |   0.872   |  0.883  |   0.931   |......|        Vase         |  0.564  |   0.698   |  0.554  |   0.742   |  0.544  |   0.752   ||       Walking       |  0.761  |   0.956   |  0.745  |   0.932   |  0.709  |   0.955   ||      Walking2       |  0.362  |   0.476   |  0.263  |   0.371   |  0.793  |   0.941   ||        Woman        |  0.615  |   0.908   |  0.648  |   0.887   |  0.771  |   0.936   |-----------------------------------------------------------------------------------------

OTB100 Success Plot	OTB100 Precision Plot

3. Evaluation on VOT2018-LT

cd /path/to/pysot-toolkit
python bin/eval.py 
	--dataset_dir /path/to/dataset/root 		# dataset path
	--dataset VOT2018-LT 				# dataset name
	--tracker_result_dir /path/to/tracker/dir 	# tracker dir
	--trackers SiamRPN++ MBMD DaSiam-LT 		# tracker names 
	--num 4 				  	# evaluation thread
	--vis 					  	# wether to draw graph# you will see-------------------------------------------|Tracker Name| Precision | Recall |  F1   |-------------------------------------------| SiamRPN++  |   0.649   | 0.610  | 0.629 ||    MBMD    |   0.634   | 0.588  | 0.610 || DaSiam-LT  |   0.627   | 0.588  | 0.607 ||    MMLT    |   0.574   | 0.521  | 0.546 ||  FuCoLoT   |   0.538   | 0.432  | 0.479 ||  SiamVGG   |   0.552   | 0.393  | 0.459 ||   SiamFC   |   0.600   | 0.334  | 0.429 |-------------------------------------------

Get Tracking Results of Your Own Tracker

Add pysot-toolkit to your PYTHONPATH

export PYTHONPATH=/path/to/pysot-toolkit:$PYTHONPATH

1. OPE (One Pass Evaluation)

from pysot.datasets import DatasetFactory

dataset = DatasetFactory.create_dataset(name=dataset_name,                                       	dataset_root=datset_root,                                        load_img=False)for video in dataset:    for idx, (img, gt_bbox) in enumerate(video):        if idx == 0:            # init your tracker here
        else:            # get tracking result here

2. Restarted Evaluation

from pysot.datasets import DatasetFactoryfrom pysot.utils.region import vot_overlap

dataset = DatasetFactory.create_dataset(name=dataset_name,                                       	dataset_root=datset_root,                                        load_img=False)
frame_counter = 0pred_bboxes = []for video in dataset:    for idx, (img, gt_bbox) in enumerate(video):        if idx == frame_counter:            # init your tracker here
            pred_bbox.append(1)        elif idx > frame_counter:            # get tracking result here
            pred_bbox = 
            overlap = vot_overlap(pred_bbox, gt_bbox, (img.shape[1], img.shape[0]))            if overlap > 0: 
	    	# continue tracking
                pred_bboxes.append(pred_bbox)            else: 
	    	# lost target, restart
                pred_bboxes.append(2)
                frame_counter = idx + 5
        else:
            pred_bboxes.append(0)