Issue about alignment between label and frames. about vac_cslr

[ycmin95]

@hulianyuyy
There are two widely used ways to align the labels with features, the greedy alignment (selecting the most likely gloss at each step) and the dominant alignment (DNF, TMM'19). We adopt the greedy alignment for simplicity and show the network predictions.
The alignment between features and frames is simply based on its temporal receptive field, for example, the temporal receptive field of Subgloss-wise conv1d (C5-P2) is 6, so the logit z_t is corresponding to frames f_[t2, t2+6] (no padding setting).

[hulianyuyy]

A question that confused me is how to find the ground truth label of frames (as marked in Fig.5 in your paper). As far as we know, the Phoenix Dataset only provide labels in video domain but not precisely assigned to frames.

[ycmin95]

@hulianyuyy
We mannually labelled several sequences for visualization. (This website may be helpful for Phoenix).

[hulianyuyy]

Many thanks for you kind reply.