I’m an undergraduate at East China University of Science and Technology and a research intern at ComPLightLab, UCL.
I study how visual representations can be made more controllable, interpretable, and reliable—from Gaussian primitives to physics-anchored imaging systems.
Source · Gaussian fit · Keyed rendering Source photograph: Christian Widell, CC0
Accepted ACM SIGGRAPH Asia 2026 · Technical Communications
Gaussian Image Steganography via Parameter-Domain Keyed Embeddings
Tong Wu, Runze Cheng, Xiaoyue Fan, Kaan Akşit
Embedding a keyed payload in the parameters of a fitted 2D Gaussian representation while preserving its rendered appearance.
Research details
I built the PyTorch fitting, rendering, and evaluation pipeline, and designed a directional embedding-cost rule combining image sensitivity, parameter plausibility, visual prominence, and detectability.
On a locked 112-image held-out set at 256 × 256 with 4,096 Gaussians, all 112 eight-bit payloads decoded exactly. Wrong-key bit error rate was 0.520, with a mean fidelity cost of 0.009 dB.
At 5% sampling, the physics-anchored model improved Fashion-MNIST ResNet accuracy by approximately 2.6 percentage points and reduced run-to-run accuracy standard deviation by approximately 9.2× across five runs.
Across nine acquisition campaigns, I collected 1,876 RAW exposures and investigated illumination, calibration, and field-transfer errors. Multi-plane measurements recovered band-limited incident phase without an interferometer.
The remaining with-array transfer was unresolved and associated with a pattern-dependent display-state change. The public toolkit preserves these limits alongside reusable simulation and validation code, with 103 tests and continuous integration.
Research experience
Mar 2026 — Present
ComPLightLab, University College London
Research Intern · With Prof. Kaan Akşit
Gaussian representations, parameter-domain embedding, and differentiable rendering.
Feb 2026 — Present
QCI Lab, Warsaw University of Technology
Undergraduate Researcher · With Dr. Mikołaj Rogalski and Prof. Maciej Trusiak · Remote
Condition-specific smoothing and validation in quantitative phase imaging.
Jun — Jul 2026
Shimobaba Laboratory, Chiba University
Research Visitor · With Prof. Tomoyoshi Shimobaba
Optical-bench experiments, calibration, and multi-plane phase retrieval. Remote from June; two weeks on site.
Education
Sep 2024 — Jul 2028
East China University of Science and Technology
B.Eng. in Optoelectronic Information Science and Engineering (expected)
Shanghai, China
Academic record
GPA 3.7627 / 4.00 · Average 90.93 / 100 · Ranked 3rd in the program, as of July 2026.
