Aleksandr Razin

Hey! 👋

I am Aleksandr Razin, an MSc graduate in Machine Learning from ITMO University, with research interests at the intersection of image restoration, generative computer vision, and multimodal learning. My long-term research objective is to develop controllable and personalized generative systems that integrate structural understanding with efficient synthesis, bridging low-level vision and modern generative modeling under practical deployment constraints.

My research experience spans both academic and industrial environments. I have contributed to pipeline-inspection robotics, resulting in a granted patent, as well as to industrial computer vision systems. Subsequently, at Huawei Research, I worked on image and video restoration and diffusion-based image generation. My contributions focused on improving perceptual quality and high-frequency detail reconstruction, including quantized fine-tuning for the Mate-70 smartphone to mitigate night-HDR ghosting and flicker, optimization of diffusion model objectives to enhance texture generation, and architectural and training strategy modifications to reduce latency in production models.

My MSc thesis investigated spatial conditioning in generative models, demonstrating that incorporating auxiliary spatial cues, such as depth maps, into super-resolution frameworks provides structural priors that alleviate common restoration and synthesis artifacts, including blur, defocus, and spatial inconsistency.

I am actively seeking PhD opportunities in generative modeling, image restoration, and multimodal learning. I am also interested in 3D generative modeling and am currently conducting research on controllable animation generation from hand-drawn sketches using ControlNet-based pose alignment.

You can find my CV here. I’m always open to collaboration, feel free to reach out by email!

Selected Publications

arXiv

One Small Step in Latent, One Giant Leap for Pixels: Fast Latent Upscale Adapter for Your Diffusion Models

Aleksandr Razin, Danil Kazantsev, and Ilya Makarov

2025

arXiv preprint arXiv:2511.10629

arXiv Bib

@misc{razin2025smallsteplatentgiant,
  title = {One Small Step in Latent, One Giant Leap for Pixels: Fast Latent Upscale Adapter for Your Diffusion Models},
  author = {Razin, Aleksandr and Kazantsev, Danil and Makarov, Ilya},
  year = {2025},
  eprint = {2511.10629},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2511.10629},
  note = {arXiv preprint arXiv:2511.10629},
}

JSTITMO

Improving question answering in programming domain with pretrained language model finetuning using structured diverse online forum data

Alexey V. Gorbatovski, Alexandr D. Razin, Artem A. Aliev, and 1 more author

Journal Scientific and Technical Of Information Technologies, Mechanics and Optics, 2024

Bib HTML

@article{gorbatovski2024improving,
  title = {Improving question answering in programming domain with pretrained language model finetuning using structured diverse online forum data},
  author = {Gorbatovski, Alexey V. and Razin, Alexandr D. and Aliev, Artem A. and Kovalchuk, Sergey V.},
  journal = {Journal Scientific and Technical Of Information Technologies, Mechanics and Optics},
  volume = {158},
  number = {6},
  pages = {1024},
  year = {2024},
}

Infrastructures

Robot crawler for surveying pipelines and metal structures of complex spatial configuration

Vladimir Pshenin, Anastasia Liagova, Alexander Razin, and 2 more authors

Infrastructures, 2022

Bib HTML

@article{pshenin2022robot,
  title = {Robot crawler for surveying pipelines and metal structures of complex spatial configuration},
  author = {Pshenin, Vladimir and Liagova, Anastasia and Razin, Alexander and Skorobogatov, Alexander and Komarovsky, Maxim},
  journal = {Infrastructures},
  volume = {7},
  number = {6},
  pages = {75},
  year = {2022},
  publisher = {MDPI},
}

RU Patent

Program for processing data for the presence of corrosion damage and its detection using a convolutional neural network

G.S. Zakirova, V.V. Pshenin, and A.D. Razin

2022

Application number: 2022615229

Bib

@misc{patent2022corrosion,
  title = {Program for processing data for the presence of corrosion damage and its detection using a convolutional neural network},
  author = {Zakirova, G.S. and Pshenin, V.V. and Razin, A.D.},
  year = {2022},
  howpublished = {Russian Patent},
  number = {2022616881},
  note = {Application number: 2022615229},
}

RU Patent

Program for object detection on raw data using a pre-trained neural network model for classification

E.R. Pshenin, V.V. Pshenin, A.D. Razin, and 1 more author

2022

Application number: 2022661008

Bib

@misc{patent2022object,
  title = {Program for object detection on raw data using a pre-trained neural network model for classification},
  author = {Pshenin, E.R. and Pshenin, V.V. and Razin, A.D. and Dzhemilev},
  year = {2022},
  howpublished = {Russian Patent},
  number = {2022662358},
  note = {Application number: 2022661008},
}

RU Patent

Database of typical structural elements of gas pipelines, collected during testing of a robotic device and processed by a convolutional neural network

E.R. Dzhemilev, V.V. Pshenin, and A.D. Razin

2022

Application number: 2022621389

Bib

@misc{patent2022database,
  title = {Database of typical structural elements of gas pipelines, collected during testing of a robotic device and processed by a convolutional neural network},
  author = {Dzhemilev, E.R. and Pshenin, V.V. and Razin, A.D.},
  year = {2022},
  howpublished = {Russian Patent},
  number = {2022621490},
  note = {Application number: 2022621389},
}