Attribute wise evaluation of datasets

[madhukiranets]

Hi, Thanks very much for the amazing repo. Was wondering if there is a way to plot graphs attribute-wise for the tracking evaluation.

[martin-danelljan]

@2006pmach Can you check this?

[2006pmach]

Hey,

I just released code to do attribute-wise evaluation in our repo. You can either generate a table or a spider/radar plot.

from pytracking.analysis.plot_results import print_results_per_attribute, plot_attributes_radar
from pytracking.evaluation import trackerlist , get_dataset_attributes

trackers = []
trackers.extend(trackerlist('dimp', 'super_dimp', range(0,5), 'Super DiMP'))

print_results_per_attribute(trackers, get_dataset_attributes('lasot', mode='short'), 'lasot',
                             merge_results=True, force_evaluation=False,
                             skip_missing_seq=True,
                             exclude_invalid_frames=False)

plot_attributes_radar(trackers,
                      get_dataset_attributes('lasot', mode='long'), 'lasot',
                      merge_results=True, force_evaluation=False,
                      skip_missing_seq=True,
                      plot_opts=None, exclude_invalid_frames=False)

Currently lasot, otb, avist and uav are supported you can choose between long (full attribute names) or short (abbreviated attribute names).

[2006pmach]

Or if you want to plot the curves for each attribute the following should work

from pytracking.analysis.plot_results import plot_results
from pytracking.evaluation import trackerlist , get_dataset_attributes

trackers = []
trackers.extend(trackerlist('dimp', 'super_dimp', range(0,5), 'Super DiMP'))

for key, dataset in get_dataset_attributes('lasot', mode='short').items():
    print(key)
    plot_results(trackers, dataset, 'lasot' + key, merge_results=True, force_evaluation=False, skip_missing_seq=False, plot_types=('success'))