Soft Facial Landmark Detection by Label Distribution Learning
Authors: Kai Su, Xin Geng5008-5015
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that the proposed method performs better than the compared state-of-the-art facial landmark detection algorithms. Furthermore, the proposed method appears to be much more robust against the landmark noise in the training set than other compared baselines. |
| Researcher Affiliation | Academia | MOE Key Laboratory of Computer Network and Information Integration, School of Computer Science and Engineering, Southeast University, Nanjing 210096, China {sukai, xgeng}@seu.edu.cn |
| Pseudocode | No | No pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | No | The paper does not provide any explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | We perform an experiment to see how our proposed MCBR method outperforms other CLM methods. The experiment is performed on the Bio ID database (Jesorsky, Kirchberg, and Frischholz 2001)... Evaluations are conducted on the 300-W dataset (Sagonas et al. 2013a)... our training set consists of the training set of LFPW (Belhumeur et al. 2013) and Helen (Le et al. 2012), the whole AFW set (Zhu and Ramanan 2012)... |
| Dataset Splits | No | Our training set consists of the training set of LFPW (Belhumeur et al. 2013) and Helen (Le et al. 2012), the whole AFW set (Zhu and Ramanan 2012), totally 3148 training images. Our test set consists of the test set of LFPW and Helen, and the whole IBUG set, totally 689 test images. |
| Hardware Specification | Yes | The average run time for our proposed MCBR method on an Intel Core i7 2.20-GHz machine is 140 ms per image with 20 landmarks. |
| Software Dependencies | No | The limited-memory quasi-Newton method L-BFGS (Liu and Nocedal 1989) is used to optimize Equation (9). Our proposed MCBR method using unoptimized Python implementations. No version numbers are provided for L-BFGS or Python or any other libraries. |
| Experiment Setup | Yes | In this experiment, we conduct our proposed MCBR method at m = 3 different resolution layers, each layer contains n = 2 stages. The standard deviation in Equation (6) to compute the BLDs is set to 2. And the size of cropped patch is set to 31 31. The number of iterations in L-BFGS is set to 66. To provide a better initial shape for an image, we divide the training set into three view-speciļ¬c subsets, i.e., left ( 30 , 0 ), frontal ( 15 , 15 ) and right (0 , 30 ). |