Abstract
Sign language recognition (SLR) has recently achieved a breakthrough inperformance thanks to deep neural networks trained on large annotated signdatasets. Of the many different sign languages, these annotated datasets areonly available for a select few. Since acquiring gloss-level labels on signlanguage videos is difficult, learning by transferring knowledge from existingannotated sources is useful for recognition in under-resourced sign languages.This study provides a publicly available cross-dataset transfer learningbenchmark from two existing public Turkish SLR datasets. We use a temporalgraph convolution-based sign language recognition approach to evaluate fivesupervised transfer learning approaches and experiment with closed-set andpartial-set cross-dataset transfer learning. Experiments demonstrate thatimprovement over finetuning based transfer learning is possible withspecialized supervised transfer learning methods.