SegmentalConsensus

Enums · AiDotNet.Safety.Video

The segmental consensus function G of a Temporal Segment Network, which aggregates per-snippet class scores into one video-level score.

For Beginners

The video is chopped into a few equal chunks and one frame is examined from each. This setting decides how those separate opinions are combined into a single verdict for the whole video — average them, take the most extreme, or weight some chunks more heavily.

How It Works

Wang et al., “Temporal Segment Networks: Towards Good Practices for Deep Action Recognition” (ECCV 2016, arXiv:1608.00859) evaluate three forms of G. On UCF101 split 1 (Table 3), spatial / temporal / two-stream accuracy:

The paper concludes “average pooling function achieves the best performance”, which is why Average is the default here.

Fields

Field Summary
SegmentalConsensus Average Even average across snippets — the paper’s best-performing consensus and the default.
SegmentalConsensus Max Maximum across snippets.
SegmentalConsensus WeightedAverage Linearly increasing weights across snippets, normalized to sum to one, so later segments count for more.
int value__