{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/empirically-analyzing-the-effect-of-dataset","title":"Empirically Analyzing the Effect of Dataset Biases on Deep Face Recognition Systems","arxiv_id":"1712.01619","date":"2017-12-05","proceeding":null,"authors":["Adam Kortylewski","Bernhard Egger","Andreas Schneider","Thomas Gerig","Andreas Morel-Forster","Thomas Vetter"],"abstract":"It is unknown what kind of biases modern in the wild face datasets have\nbecause of their lack of annotation. A direct consequence of this is that total\nrecognition rates alone only provide limited insight about the generalization\nability of a Deep Convolutional Neural Networks (DCNNs). We propose to\nempirically study the effect of different types of dataset biases on the\ngeneralization ability of DCNNs. Using synthetically generated face images, we\nstudy the face recognition rate as a function of interpretable parameters such\nas face pose and light. The proposed method allows valuable details about the\ngeneralization performance of different DCNN architectures to be observed and\ncompared. In our experiments, we find that: 1) Indeed, dataset bias has a\nsignificant influence on the generalization performance of DCNNs. 2) DCNNs can\ngeneralize surprisingly well to unseen illumination conditions and large\nsampling gaps in the pose variation. 3) Using the presented methodology we\nreveal that the VGG-16 architecture outperforms the AlexNet architecture at\nface recognition tasks because it can much better generalize to unseen face\nposes, although it has significantly more parameters. 4) We uncover a main\nlimitation of current DCNN architectures, which is the difficulty to generalize\nwhen different identities to not share the same pose variation. 5) We\ndemonstrate that our findings on synthetic data also apply when learning from\nreal-world data. Our face image generator is publicly available to enable the\ncommunity to benchmark other DCNN architectures.","url_abs":"http://arxiv.org/abs/1712.01619v4","url_pdf":"http://arxiv.org/pdf/1712.01619v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"empirically-analyzing-the-effect-of-dataset","repo_url":"https://github.com/unibas-gravis/parametric-face-image-generator","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"empirically-analyzing-the-effect-of-dataset","repo_url":"https://github.com/Arneli/image-generator-for-BScThesis","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"face-recognition","task_name":"Face Recognition"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dcnn","method_name":"DCNN"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"local-response-normalization","method_name":"Local Response Normalization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1712.01619","atlas_url":"https://app.syntology.ai/?focus=1712.01619","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}