Deep Learning Face Representation by Joint Identification-Verification
Authors: Yi Sun, Yuheng Chen, Xiaogang Wang, Xiaoou Tang
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | On the challenging LFW dataset [11], 99.15% face verification accuracy is achieved. Compared with the best previous deep learning result [20] on LFW, the error rate has been significantly reduced by 67%. |
| Researcher Affiliation | Collaboration | 1Department of Information Engineering, The Chinese University of Hong Kong 2Sense Time Group 3Department of Electronic Engineering, The Chinese University of Hong Kong 4Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences |
| Pseudocode | Yes | Table 1: The Deep ID2 feature learning algorithm. |
| Open Source Code | No | No explicit statement about providing open-source code for the described methodology or a link to a code repository was found. |
| Open Datasets | Yes | We report face verification results on the LFW dataset [11]... In particular, we use the Celeb Faces+ dataset [20] for training, which contains 202,599 face images of 10,177 identities (celebrities) collected from the Internet. |
| Dataset Splits | Yes | Deep ID2 features are learned from the face images of 8192 identities randomly sampled from Celeb Faces+ (referred to as Celeb Faces+A), while the remaining face images of 1985 identities (referred to as Celeb Faces+B) are used for the following feature selection and learning the face verification models (Joint Bayesian). When learning Deep ID2 features on Celeb Faces+A, Celeb Faces+B is used as a validation set to decide the learning rate, training epochs, and hyperparameter λ. After that, Celeb Faces+B is separated into a training set of 1485 identities and a validation set of 500 identities for feature selection. |
| Hardware Specification | Yes | The feature extraction process is also efficient and takes only 35 ms for each face image with a single Titan GPU. |
| Software Dependencies | No | The paper mentions specific algorithms and models (e.g., 'rectified linear units (Re LU) [17]', 'Joint Bayesian model [3]', 'SDM algorithm [24]') but does not provide specific version numbers for any software packages, libraries, or dependencies. |
| Experiment Setup | Yes | The margin m in Eq. (2) is a special case, which cannot be updated by gradient descent since this will collapse it to zero. Instead, m is fixed and updated every N training pairs (N 200, 000 in our experiments) such that it is the threshold of the feature distances fi fj to minimize the verification error of the previous N training pairs. Updating m is not included in Tab. 1 for simplicity. |