csl-daily dataset training

[jekim5418]

Thank you for your research.

I found that there exist experiments of PHOENIX-2014T and CSL-daily and the model architecture is based on Sign2(Gloss+Text).
The sign2 (Gloss+Text) model uses features generated by pretrained spatial embeddings, but there is no feature pickle in the csl-daily dataset. It would be appreciated if you could tell us how the features created by csl-daily were obtained.

[biaofuxmu]

Thank you for your research.

I found that there exist experiments of PHOENIX-2014T and CSL-daily and the model architecture is based on Sign2(Gloss+Text). The sign2 (Gloss+Text) model uses features generated by pretrained spatial embeddings, but there is no feature pickle in the csl-daily dataset. It would be appreciated if you could tell us how the features created by csl-daily were obtained.

We use the features from the file Extracted_Features_with_BN-TIN-Transformer.zip in the CSL-Daily dataset.