Dense Temporal Convolution Network for Sign Language Translation
Authors: Dan Guo, Shuo Wang, Qi Tian, Meng Wang
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results on two popular sign language benchmarks, i.e. PHOENIX and USTCCon Sents, demonstrate the effectiveness of our proposed method in terms of various measurements. |
| Researcher Affiliation | Collaboration | 1School of Computer Science and Information Engineering, Hefei University of Technology 2Huawei Noah s Ark Lab Department of Computer Science, University of Texas at San Antonio |
| Pseudocode | No | The paper does not include pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link for the open-source code of the described methodology. |
| Open Datasets | Yes | We evaluate our method on two benchmarks: German continuous sign language dataset (PHOENIX)1 and Chinese sign language dataset (USTC-Con Sents)2. 1https://www-i6.informatik.rwth-aachen.de/ koller/RWTHPHOENIX/ 2http://mccipc.ustc.edu.cn/mediawiki/index.php/SLR Dataset |
| Dataset Splits | Yes | In the PHOENIX dataset, we split each video into 8-frames and overlapped by 4-frames. Therefore, we acquire 190536 / 17908 / 21349 clips from TRAIN / VAL / TEST sets, respectively. [...] (a) Split I is a signer independent test. The TRAIN set contains samples of 40 signers and the remaining of 10 signers as the TEST set. The sentences of the TEST set are existing in the TRAIN set. (b) Split II is an unseen sentences test which selects video samples of 94 sentences as the TRAIN set and the remaining 6 as the TEST set. |
| Hardware Specification | No | The paper mentions calculations are limited by 'calculation capacity of GPU', but does not provide specific hardware details such as GPU model, CPU type, or memory. |
| Software Dependencies | No | The paper mentions using ReLU, CTC, and ADAM optimization, but does not provide specific version numbers for any software libraries or dependencies (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | In the training stage, we use Re LU as activation function [Krizhevsky et al., 2012] and the parameter of the dropout is ρ = 0.5. Then, we train our network by CTC object function in ADAM optimization starting with the learning rate of 10 4, beats range from 0.5 to 0.999, and weight decay is 10 5. We reduce the learning rate by 0.1 after each 30 training epoch and stop the training stage when the learning rate is lower than 10 6. |