Portrait of Khoa Pham-Dinh

Khoa Pham-Dinh

Research Intern, Trustworthy and Adversarial Computing Lab, Aalto University

Helsinki, Finland

About

I am a research intern at the Trustworthy and Adversarial Computing Lab at Aalto University, working on adversarial examples for cyber-physical vision systems: attacks realized in the physical world — printed patches, modified objects, AR/VR — rather than purely digital perturbations. The work is gradient-based optimization of attacks on deep computer vision models under physical constraints.

Before that I spent a year with the Visual AI System Research team at Nokia Technologies and Aalto University, on super-resolution post-filters for Versatile Video Coding, low-rank adaptation for finetuning, and quantization-aware training. My MSc thesis, written at the Signal Analysis and Machine Intelligence group in Tampere, recast joint compression–segmentation training as constrained optimization over the rate–quality–performance trade-off.

Parts of this work have gone into standardization: two contributions to MPEG and JVET, one of which was adopted into the Common Test Conditions used to develop the Video Coding for Machines standard.

News

  • Sep 2026 CASR received the Best Student Paper Award at the IEEE 28th International Workshop on Multimedia Signal Processing (MMSP 2026) in Istanbul.

Publications

Peer-reviewed

  1. CASR: Content-Adaptive Neural Super-Resolution Post-Filter for Versatile Video Coding via Low-Rank Overfitting

    Khoa Pham-Dinh, Francesco Cricri, Maria Santamaria, Honglei Zhang, Hamed Rezazadegan Tavakoli, Moncef Gabbouj, Juho Kannala, Miska M. Hannuksela

    IEEE International Workshop on Multimedia Signal Processing (MMSP), Istanbul, Turkey, 22–24 September 2026.

  2. Evaluating the Emerging MPEG Video Coding for Machines in Semantic Segmentation

    Khoa Pham-Dinh, Farhad Pakdaman, Honglei Zhang, Hamed Rezazadegan Tavakoli, Nam Le, Jukka Ahonen, Moncef Gabbouj

    IEEE International Symposium on Multimedia (ISM), 2025, pp. 141–148. Candidate for Best Paper Award.

  3. Estimating Intra- and Inter-Subject Oxygen Consumption in Outdoor Human Gait Using Multiple Neural Network Approaches

    Philipp Müller†, Khoa Pham-Dinh†, Huy Trinh, Anton Rauhameri, Neil J. Cronin

    PLOS ONE 19(9):e0303317, 2024. † equal contribution.

Preprint

  1. You’ve Seen Enough: Quality-Constrained Image Coding For Machines

    Khoa Pham-Dinh, Sanaz Nami, Hamed Rezazadegan Tavakoli, Moncef Gabbouj, Farhad Pakdaman

Standardization contributions

  • S2.

    Neural Network Super Resolution as Postfilter with Content Adaptation

    Document JVET-AP0178-v1, Joint Video Experts Team, 42nd meeting, April 2026. ITU-T SG21 WP3/21 and ISO/IEC JTC 1/SC 29.

  • S1.

    Semantic Segmentation for VCM with PandaSet

    Document m67677, ISO/IEC JTC 1/SC 29/WG 4 (MPEG Video Coding), April 2024. Adopted into the Common Test Conditions used to develop the Video Coding for Machines standard.