Report generation

This tutorial shows how to generate reports from tracker results in a workspace.

Goal

At the end, you will be able to generate:

  • an HTML report,

  • a LaTeX report,

  • plot/video exports.

Prerequisites

Before generating a report, make sure you have:

  • an initialized workspace,

  • at least one integrated tracker,

  • evaluation results (run vot evaluate).

Step 1: Run evaluation

Generate tracker outputs first:

vot evaluate --workspace ./workspace mytracker

You can provide multiple trackers:

vot evaluate --workspace ./workspace tracker_a tracker_b

Step 2: Generate a basic HTML report

Create a default report:

vot report --workspace ./workspace --format html mytracker

If –name is not provided, the toolkit creates a timestamp-based report name.

Step 3: Choose report format

The command supports three formats:

  • html - interactive browser report,

  • latex - LaTeX document output,

  • plots - exported plot/video files only.

Examples:

vot report --workspace ./workspace --format latex --name paper_report mytracker
vot report --workspace ./workspace --format plots --name figures mytracker

Step 4: Filter report content

You can restrict report generation to selected experiments and sequences.

vot report --workspace ./workspace \
  --experiments baseline,realtime \
  --sequences bike1,bike2 \
  --format html \
  --name focused_report \
  mytracker

This is useful for faster iteration when debugging integrations or preparing a concise comparison.

Step 5: Compare multiple trackers

Generate one report with multiple trackers:

vot report --workspace ./workspace --format html tracker_a tracker_b tracker_c

The toolkit sorts trackers according to report configuration and places all selected trackers in one output document.

Step 6: Customize reports in config.yaml

Workspace report behavior is controlled by the report section in config.yaml.

The available configuration blocks are:

  • report.style - plot/axes/figure callables,

  • report.sort - tracker ordering before report generation,

  • report.index - list of report generators to include.

Minimal custom index (equivalent to default overview):

report:
  index:
    - type: table
    - type: plots

You can also include additional built-in report generators:

report:
  index:
    - type: table
    - type: plots
    - type: vot.report.common.SequenceOverlapPlots
      ignore_masks: _ignore

For a description of each element see Available report elements below.

Sort trackers by a selected analysis output before plots/tables are produced:

report:
  sort:
    experiment: baseline
    analysis: <analysis-name>
    result: 0

Notes:

  • experiment must match experiment identifier from the stack (for example baseline).

  • analysis must match Analysis.name (not necessarily class name).

  • result is a zero-based index into analysis outputs returned by describe().

Define local style hooks in your workspace (for example report_hooks.py):

from matplotlib.axes import Axes
from vot.report import DefaultStyle

class BasicStyle(DefaultStyle):
    def line_style(self, opacity=1):
        style = super().line_style(opacity=opacity)
        style["linewidth"] = 2
        return style

    def point_style(self):
        style = super().point_style()
        style["s"] = 40
        return style

def basic_plot_style(index):
    return BasicStyle(index)

def basic_axes(figure, rect=None, _=None):
    axes = Axes(figure, rect or [0, 0, 1, 1])
    axes.grid(visible=True, color=[0.85, 0.85, 0.85], linestyle="--", linewidth=1)
    figure.add_axes(axes)
    return axes

Reference these hooks from config.yaml:

report:
  style:
    plots: report_hooks.basic_plot_style
    axes: report_hooks.basic_axes

Explicit tracker-ordering example in config.yaml:

report:
  sort:
    experiment: baseline
    analysis: Quality
    result: 0

The analysis value above must match an analysis name that exists in your stack. If your stack defines a different name, replace Quality accordingly.

After editing config.yaml, regenerate the report:

vot report --workspace ./workspace --format html tracker_a tracker_b

Available report elements

All report elements are specified under report.index in config.yaml using the type key. Two short aliases are built in; all others must use the full Python class path.

table (vot.report.common.StackAnalysesTable)

Produces an overview table of all per-tracker scalar measures computed by the analyses defined in the stack. Each row is a tracker; each column group corresponds to one experiment/analysis combination. Columns are ranked so that the best result in each column is highlighted. This element has no parameters.

- type: table
plots (vot.report.common.StackAnalysesPlots)

Produces one plot per analysis output (scatter plots for 2-D point results, line plots for curve/plot results) for every experiment in the stack. All trackers are drawn in reverse score order so that the best-performing tracker is rendered on top. This element has no parameters.

- type: plots
SequenceOverlapPlots (vot.report.common.SequenceOverlapPlots)

Produces one overlap-over-time line plot per sequence for every experiment. Each line represents one tracker run. This element is only compatible with multi-run experiments (MultiRunExperiment); it is silently skipped for other experiment types.

Parameters:

  • ignore_masks (string, default ``_ignore``) — object identifier used to fetch per-frame ignore masks from the dataset (regions marked with this label are excluded from the overlap calculation).

- type: vot.report.common.SequenceOverlapPlots
  ignore_masks: _ignore
PreviewVideos (vot.report.video.PreviewVideos)

Produces rendered sequence preview videos with tracker trajectories overlaid on frames. This element is only compatible with multi-run experiments (MultiRunExperiment); it is silently skipped for other experiment types.

Parameters:

  • groundtruth (boolean, default ``false``) - include ground-truth regions

    in the preview.

  • separate (boolean, default ``false``) - request separate videos per

    tracker. In current toolkit code, prefer the default combined mode.

- type: vot.report.video.PreviewVideos
  groundtruth: true
  separate: false

Where outputs are stored

Report outputs are stored in:

<workspace>/reports/<report-name>/

Typical files include:

  • report.html for HTML format,

  • report.tex for LaTeX format,

  • *.png, *.pdf, *.avi exports for plots format.

Notes on report content

By default, report content is based on report configuration in the workspace. If no custom report configuration is present, the toolkit generates a default overview (analysis table and analysis plots).

Troubleshooting

  • No trackers resolved: verify tracker identifier in trackers.ini and command arguments.

  • No experiments/sequences selected: check names used in –experiments and –sequences.

  • Empty report: ensure evaluation has completed for the selected tracker(s).

  • Missing dependencies for rendering: start with –format html (most robust default).