Sign language subtitle alignment on BOBSL (test)
0.6568F1@0.50SignCLIP-BSL + SAT+ subtitles
Evaluation Results
| Method | Links | |
|---|---|---|
| SignCLIP-BSL + SAT+ subtitlesAlignment Approach=Segment, Embed, and Align, Strategy=initialized from SAT+ subtitles2025.12 | 0.6568 | |
| SAT+Model Type=End-to-end trained2025.12 | 0.6381 | |
| SATModel Type=End-to-end trained2025.12 | 0.5457 | |
| SignCLIP-BSL/ASL/DSGS-finetunedAlignment Approach=Segment, Embed, and Align, Strategy=finetuned2025.12 | 0.545 | |
| SignCLIP-multilingualAlignment Approach=Segment, Embed, and Align, Strategy=multilingual2025.12 | 0.5068 | |
| SegmentationAlignment Approach=Segment and Align2025.12 | 0.4958 | |
| Segmentation-BSLCPAlignment Approach=Segment and Align2025.12 | 0.4775 | |
| Original+ (fixed offsets from training data)Alignment Approach=Fixed offsets2025.12 | 0.4461 | |
| Original alignment (usually audio-based)Alignment Approach=Audio-based2025.12 | 0.1411 |