Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Sign-IDD: Iconicity Disentangled Diffusion for Sign Language Production

Authors: Shengeng Tang, Jiayi He, Dan Guo, Yanyan Wei, Feng Li, Richang Hong

AAAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on PHOENIX14T and USTC-CSL datasets validate the effectiveness of our method. [...] Experiments are conducted using Py Torch on NVIDIA Ge Force RTX 2080 Ti GPUs.
Researcher Affiliation Academia Shengeng Tang, Jiayi He, Dan Guo, Yanyan Wei*, Feng Li, Richang Hong* School of Computer Science and Information Engineering, Hefei University of Technology EMAIL, EMAIL
Pseudocode No The paper describes methods using mathematical formulations and descriptive text, but no explicit pseudocode or algorithm blocks are provided.
Open Source Code Yes Code https://github.com/Na Vi-start/Sign-IDD
Open Datasets Yes We evaluate the proposed method on two benchmarks: PHOENIX14T (Camgoz et al. 2018) and USTC-CSL (Huang et al. 2018).
Dataset Splits Yes USTC-CSL encompasses 100 Chinese sign language sentences performed by 50 signers and is divided into 4,000 training instances and 1,000 testing instances (Guo et al. 2018). [...] PHOENIX14T Table 1 provides comparison results of the proposed Sign-IDD with other SOTA methods on PHOENIX14T. As shown in this table, Sign-IDD significantly outperforms other non-diffusion-based approaches, achieving 25.40% and 24.80% BLEU-1 on the DEV and TEST sets, respectively.
Hardware Specification Yes Experiments are conducted using Py Torch on NVIDIA Ge Force RTX 2080 Ti GPUs.
Software Dependencies No Experiments are conducted using Py Torch on NVIDIA Ge Force RTX 2080 Ti GPUs. The paper mentions PyTorch but does not specify a version number.
Experiment Setup Yes The Transformer-based Gloss Encoder is built with 2 layers, 4 heads, and an embedding size of 512. In addition, we set the timesteps t of the diffusion model to 1000 and the number of inferences i to 5. During training, we use the Adam optimizer (Kingma and Ba 2015) and a learning rate of 1 10 3.