Researches | Aylin Naebzadeh

Wed, 12 Nov 2025 00:00:00 +0000

3D Reconstruction and Neural Rendering

Many real-world tasks require understanding scenes in 3D from sparse or unconstrained inputs. I work on reconstructing 3D scenes using COLMAP for camera pose estimation, followed by neural rendering methods such as NeRF and Gaussian Splatting for high-fidelity reconstruction and view synthesis. My work includes integrating object detection and spatial reasoning into these pipelines, enabling more accurate and semantically aware 3D representations.

Wed, 12 Nov 2025 00:00:00 +0000

Diffusion-Based Image Restoration

Real-world imaging systems introduce complex degradations that standard diffusion models are not designed to handle. My work focuses on developing diffusion-based restoration methods that remain reliable under realistic sensor conditions, including spatially correlated noise, low resolution, and thermal imaging artifacts. I explore principled transformations such as whitening to convert correlated sensor noise into forms compatible with diffusion sampling, enabling accurate restoration without retraining. I also design patch-based diffusion frameworks that learn local thermal priors from small image regions and use overlapping patch denoising with smooth blending to reconstruct full-resolution outputs. Across denoising, super-resolution, and deblurring, these approaches improve robustness and generalization for both RGB and thermal images and aim to make diffusion-based restoration practical for real-world deployment.

Wed, 12 Nov 2025 00:00:00 +0000

Robust Multimodal Learning

Multimodal systems often struggle when one or more input sources are missing, corrupted, or imbalanced. My work develops learning strategies that preserve performance under these real-world conditions by enabling models to reason effectively with incomplete information. This includes designing projection-based methods that learn to estimate missing modality representations directly from the available ones and training unified models that handle all missing-modality configurations without the need for modality-specific retraining or complex adaptation procedures. I also explore efficient fine-tuning techniques, such as LoRA-based unimodal adaptation, to reduce computational overhead while maintaining strong performance across diverse modalities and tasks. These efforts benefit from my background in NLP, which strengthens my understanding of cross-modal interactions and how language models can support robust reasoning when some modalities are absent.

Wed, 12 Nov 2025 00:00:00 +0000

Research

My research spans computer vision, NLP, multimodal learning, and 3D reconstruction, with a focus on building reliable ML systems that operate robustly under real-world conditions. I develop diffusion-based methods for image restoration that handle real sensor noise, low-quality inputs, and cross-domain degradations. I also work on improving the robustness of multimodal models when information is missing, corrupted, or imbalanced across modalities. Beyond 2D vision, I explore 3D reconstruction and neural rendering, including camera pose estimation, NeRF, and Gaussian Splatting. More broadly, my work connects generative modeling, visual understanding, and trustworthy AI, strengthened by my background in natural language processing.

Diffusion Models for Inverse Problems

Robust Multimodal Learning

3D Reconstruction

Resources

Mon, 01 Jan 0001 00:00:00 +0000