This file shows the parameters used to generate the temporal proposals used in our preprint-v1.
The parameter below should be used to instantiate the classes here. Using those and our train/val/test data should regenerate our temporal proposals.
TODO:sample-code-to-dump-temporal-proposals per-video
{
"proposal_interface": "DidemoICCV17SS",
}
For CAL models, we enforce that a moment must have more than 2 clips by using pooled features with clip-length of 2.5 seconds.
For MCN models, pooled features with clip-length of 5s achieves slightly better performance. Thus, we used&reported those.
{
"length": 3,
"scales": [
2,
3,
4,
5,
6,
7,
8
],
"stride": 0.3,
"proposal_interface": "SlidingWindowMSRSS",
}
{
"length": 5,
"scales": [
2,
4,
6,
8,
10,
12,
14,
16,
18,
20,
22,
24,
26
],
"stride": 0.3,
"proposal_interface": "SlidingWindowMSRSS",
}