GestureDet: Real-time Student Gesture Analysis with Multi-dimensional Attention-based Detector
Authors: Rui Zheng, Fei Jiang, Ruimin Shen
IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the effectiveness and versatility of Gesture Det, which achieves 75.2% m AP on real student behavior dataset, and 74.5% on public PASCAL VOC dataset at 20fps on embedding device Nvidia Jetson TX2. To demonstrate the effectiveness and versatility of our proposed Gesture Det, we conduct extensive experiments both on our student behavior dataset and the public PASCAL VOC dataset. |
| Researcher Affiliation | Academia | Rui Zheng , Fei Jiang and Ruimin Shen Department of Computer Science and Engineering, Shanghai Jiao Tong University, China zhengr, jiangf, rmshen@sjtu.edu.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. It provides architectural diagrams but no step-by-step algorithmic descriptions. |
| Open Source Code | No | The paper does not provide any explicit statement about releasing open-source code for the described methodology, nor does it include a link to a code repository. |
| Open Datasets | Yes | PASCAL VOC [Everingham et al., 2010] dataset consists of natural images drawn from 20 classes. The detectors are trained on the union set of VOC 2007 trainval and VOC 2012 trainval, and tested on VOC 2007 test. |
| Dataset Splits | Yes | We use 29k images (out of 40k images in total) for training, then validate performances on the rest 11k subset. |
| Hardware Specification | Yes | Note that, we do not use other tricks like multi-scale training and better postprocessing like Soft-NMS. What s more, we use network inference time (without pre-processing and post-processing) tested on GPU (1080Ti), CPU (Intel i7-8700K) and embedding devices (Jetson TX2) to evaluate the inference speed of models and FLOPs (floating point operations) to evaluate the computation cost of models. |
| Software Dependencies | No | The paper states: |
| Experiment Setup | Yes | We implement on Py Torch library. The detectors are trained end-to-end on one GPU using SGD with a weight decay of 0.0001 and a momentum of 0.9, following the settings in prior works. The input image resolution is 300 300 pixels for efficiency and the batch size is set to 64 images. The data augmentation strategies are the same as the original SSD [Liu et al., 2016]. The learning rate starts from 0.002 with warm-up epochs and decays exponentially every step. |