RenzKa/sign-segmentation

Finetuing pre-trained I3D model

Closed this issue · 1 comments

In training, is MS-TCN model only trained with pre-extracted features?

And is not I3D model fine-tuned?

Hi,

Yes we train the MS-TCN model with pre-extracted I3D features.

We fine-tune the I3D model with the BSL Corpus data. You can also see our analysis about fine-tuning on different datasets in our ICASSP paper (Table 2 and Table 5).