Reference Action Segmentation Evaluation Code

This repository contains the reference code for action segmentation evaluation.

If you have a bug-fix/improvement or if you want to add a new features please send a pull request or open an issue.

Installation

The actseg library is available on PyPI.

pip install actseg

Development

make init
make test

Example Usage

All the metrics have the same api.

from actseg.eval import MoFAccuracy, Edit

pred1 = [0, 0, 0, 1, 0, 1, 1, 1, 0]
pred2 = [1, 2, 3, 0, 0, 1, 2, 3, 0, 0, 0, 1, 2, 3, 0, 0, 0, 0]

target1 = [0, 0, 1, 1, 2, 1, 1, 0, 0]
target2 = [1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3]

metrics = [MoFAccuracy(), Edit()]
for p, t in zip([pred1, pred2], [target1, target2]):
    for m in metrics:
        m(targets=t, predictions=p)

for m in metrics:
    print(m)

# MoF: 0.3333333333333333
# Edit: 52.5

Metrics

Frame-wise Metrics

1. MoF (Accuracy)
2. F1Score
3. IoD
4. IoU

Segment-wise Metrics

1. Edit (Edit distance or matching score)

Specifying Ignore Class

For some Metrics it is possible to specify the indices of classes to ignore (e.g. Background) by passing ignore_ids parameter to the constructor.

Acknowledgement

Please see src/actseg/external for external sources used in this project.