activity-plots

Plotting tools for temporal action localization (or per-frame classification) labels and predictions.

Install

pip install activity-plots

Usage

Create a data container then pass it into one of the plotting functions.

import activity_plots
import matplotlib.pyplot as plt

data = activity_plots.ActivitySegments(
    starts=[5, 13, 29], ends=[15, 17, 45], label_ids=[0, 1, 0],
    label_set=["Activity A", "Activity B"]
)
fig, ax = plt.subplots()
ax = activity_plots.multi_bar(ax, data, title="Test Plot", xlabel="Frame #")
plt.show()

Plot Types

1. multi_bar

2. continuous_shaded
   • Use to visualize probabilities (model output)

3. single_bar
   • Use when there can only be one label per frame
   • Useful to visualize multiple videos/instances

Data Containers

ActivitySegments

• Build from a set of variable-length segments

ActivityProbabilities

• Used for visualizing predictions (per-frame distributions over all activities)

segments_from_sequence

• Helper function to build ActivitySegments from a sequence of label IDs.
• Example:

seq = [1, 1, 1, 0, 2, 3, 3, 3, 4, -1, -1, 5]  # -1 is ignored ("background")
data = activity_plots.segments_from_sequence(seq)

TODO

• Github action to publish on pypi
• Stackplot
  • Check different options for "baseline"
• Ridge plot
  • Fill color with same colormap as multi-bar
• Better handling of overlapping segments
  • Could use additional offset
  • Might need to increase the y-axis range too
• Create tests