William Thong

I am a Machine Learning Engineer at Apple Zurich, with a focus on data curation, model evaluation and generative models in computer vision. Previously, I was a Research Scientist at Sony AI Zurich, where I managed the local team in Responsible AI.

I have led research projects, and helped to scale research teams, in computer vision and related fields. My research outcomes have been published at top-tier research venues (e.g., Nature, NeurIPS, ICCV), and have been covered in MIT Tech Review, The Verge, Wired, among others.

Prior to that, I did my PhD at the Video & Image Sense lab of the University of Amsterdam, under the supervision of Cees Snoek. My PhD dissertation involved visual similarity, learning with limited labels, and model biases. I was fortunate to be partially funded by multiple scholarships from NSERC.

[LinkedIn] [Google Scholar] [Github]

Conference & Journal publicatons

	Fair human-centric image dataset for ethical AI benchmarking Alice Xiang, Jerone Andrews, Rebecca Bourke, William Thong, Julienne LaChance, Tiffany Georgievski, Apostolos Modas, Aida Rahmattalabbi, Yunhao Ba, Shruti Nagpal, Orestis Papakyriakopoulos, Dora Zhao, Jinru Xue, Victoria Matthews, Linxia Gong, Austin Hoag, Mircea Cimpoi, Swami Sankaranarayanan, Wiebke Hutiri, Morgan Scheuerman, Albert Abedi, Peter Stone, Peter Wurman, Hiroaki Kitano, Michael Spranger Nature, 2025 [paper] [project page] We operationalize ethical standards for data collection and benchmark the fairness of human-centric computer vision models. The paper has been featured on the cover of Nature!
	Ethical Considerations for Responsible Data Curation Jerone Andrews, Dora Zhao, William Thong, Apostolos Modas, Orestis Papakyriakopoulos, Alice Xiang Neural Information Processing Systems Datasets and Benchmarks (NeurIPS D&B), 2023 [paper] [arxiv] [code] We lay out considerations and recommendations for responsible curation of computer vision datasets.
	Beyond Skin Tone: A Multidimensional Measure of Apparent Skin Color William Thong, Przemyslaw Joniak, Alice Xiang International Conference on Computer Vision (ICCV), 2023 [paper] [arxiv] [code] We measure apparent skin color, beyond a unidimensional scale with the luminance and hue angle.
	Augmented Datasheets for Speech Datasets and Ethical Decision-Making Orestis Papakyriakopoulos, Anna Seo Gyeong Choi, Jerone Andrews, Rebecca Bourke, William Thong, Dora Zhao, Alice Xiang, Allison Koenecke ACM Conference on Fairness, Accountability, and Transparency (FAccT), 2023 [paper] [arxiv] [code] We augment datasheets to encourage ethical considerations in speech datasets.
	Content-Diverse Comparisons improve IQA William Thong, Jose-Costa Pereira, Sarah Parisot, Ales Leonardis, Steven McDonagh British Machine Vision Conference (BMVC), 2022 [paper] [arxiv] [code] We enrich image comparisons for learning a perceptual quality metric.
	Diversely-Supervised Visual Product Search William Thong and Cees G. M. Snoek ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2022 [paper] [arxiv] [code] We create a diverse set of labels from instance, attribute and category similarities for visual product search.
	Feature and Label Embedding Spaces Matter in Addressing Image Classifier Bias William Thong, Cees G. M. Snoek British Machine Vision Conference (BMVC), 2021 [paper] [arxiv] [code] We identify and mitigate biases in both feature and label embedding spaces in image classifiers.
	Object Priors for Classifying and Localizing Unseen Actions Pascal Mettes, William Thong, Cees G. M. Snoek International Journal of Computer Vision (IJCV), 2021 [paper] [arxiv] [code] We derive spatial and semantic priors to recognize unseen actions in videos with zero training sample.
	Bias-Awareness for Zero-Shot Learning the Seen and Unseen William Thong and Cees G. M. Snoek British Machine Vision Conference (BMVC), 2020 [paper] [arxiv] [code] [video] We mitigate the classifier bias towards classes seen during training in generalized zero-shot learning.
	Open Cross-Domain Visual Search William Thong, Pascal Mettes, Cees G.M. Snoek Computer Vision and Image Understanding (CVIU), 2020 [paper] [arxiv] [code] We search for categories from any source domain to any target domain in a common semantic space.
	A Layer-Based Sequential Framework for Scene Generation with GANs Mehmet O. Turkoglu, William Thong, Luuk Spreeuwers, Berkay Kicanaoglu AAAI Conference on Artificial Intelligence (AAAI), 2019 [paper] [arxiv] [poster] [code] We compose a scene layer-by-layer, with an explicit control over the generation of all scene elements.
	Convolutional Networks for Kidney Segmentation in Contrast-Enhanced CT Scans William Thong, Samuel Kadoury, Nicolas Piché, Christopher J. Pal CMBBE: Imaging & Visualization, 2018 [paper] – initially presented at MICCAI-DLMIA 2015 We segment healthy and abnormal kidneys in CT scans with a patch-based ConvNet.
	Three-dimensional Morphology Study of Surgical Adolescent Idiopathic Scoliosis Patient from Encoded Geometric Models William Thong, Stefan Parent, James Wu, Carl-Éric Aubin, Hubert Labelle, Samuel Kadoury European Spine Journal (ESJ), 2016 [paper] We cluster scoliotic spine deformations in 3D representations with a stacked auto-encoder.
	Automatic Labeling of Vertebral Levels using a Robust Template-Based Approach Eugénie Ullmann, Jean François Pelletier Paquette, William Thong*, Julien Cohen-Adad International Journal of Biomedical Imaging (IJBI)*, 2014 [paper] We build a template to predict vertebral levels in MRI images.

Workshop & Abstract publications

A case study in fairness evaluation: Current limitations and challenges for human pose estimation
LaChance* and Thong* et al. AAAI workshops, 2023.

NTIRE 2021 challenge on perceptual image quality assessment
Gu et al. CVPR workshops, 2021.

Interactive Exploration of Journalistic Video Footage through Multimodal Semantic Matching
Ibrahimi et al. ACM Multimedia (Demo track), 2019.

Stacked Auto-Encoders for Classification of 3D Spine Models in Adolescent Idiopathic Scoliosis
Thong et al. MICCAI-CSI workshop, 2014.

Spinal Cord Toolbox: an Open-Source Framework for Processing Spinal Cord MRI Data
Cohen-Adad et al. OHBM, 2014.

Academic service

Reviewer for CVPR, ECCV, ICCV, NeurIPS.

Outstanding reviewer awards at CVPR'21 and BMVC'20 and 21.

Webpage template from Jon Barron.