Skip to content
View MDSALMANSHAMS's full-sized avatar
🎯
Focusing
🎯
Focusing
5 followers · 2 following

Achievements

Achievement: Pull Shark

Achievements

Achievement: Pull Shark

Block or report MDSALMANSHAMS

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
MDSALMANSHAMS/README.md

Md Salman Shams

Deep learning engineer — computer vision and medical imaging, and the runtime underneath both.

I work at the layer where models meet hardware: input validation, bounds and integer-overflow bugs, and CUDA kernel memory safety in PyTorch core — alongside applied CV for medical and real-world video.

upstream merges: 7 PyTorch: 3 merged OpenCV: 2 merged MONAI: 1 merged CVE fixed: CVE-2026-8461

Upstream contributions

Major open-source libraries

Project PR Contribution
pytorch/pytorch #190005 Fixed an out-of-bounds read in the batch_norm_gather_stats CUDA kernel — C++/CUDA, verified with compute-sanitizer
pytorch/pytorch #188328 Added output_padding and bias validation to the transposed-convolution meta path, so invalid shapes fail with a clear error instead of reaching the kernel
pytorch/pytorch #187820 Documented factory kwargs for torch.normal
opencv/opencv-python #1249 Shipped the FFmpeg 8.1.2 bump fixing CVE-2026-8461 (CVSS 8.8) across 7 build images
opencv/opencv-python #1233 Brought the i686 manylinux build onto a supported FFmpeg
Project-MONAI/MONAI #8945 Implemented the AbsoluteVolumeDifferenceMetric for medical image segmentation

In review: pytorch/pytorch#188327 (nn.Conv2d error inputs) · opencv/opencv-python#1255 (CVE fix on the 5.x branch)

Research codebases

Project PR Contribution
MCG-NJU/SAM2-Plus #5 Built an interactive GUI for video object segmentation

In review: rmaphoh/RETFound#55 (extending the retinal foundation model to OCT drusen segmentation)

Internal work: MedVisBonn/EyeLab — a multi-modal annotation tool for retinal imaging data. Fixed bugs and added functionality on top of the open-source codebase for production use. Changes were kept in-house and not upstreamed, per organization and client requirements.

What I focus on

  • Memory safety in tensor kernels — OOB reads, integer overflow in index math, missing shape and dtype guards. Turning sanitizer crashes into clear error messages.
  • Computer vision — detection, segmentation, tracking, OCR; deepfake and face analysis.
  • Medical imaging — retinal and volumetric segmentation, metrics, MONAI.
  • Build and supply-chain hygiene — CVE remediation in packaging pipelines.

Toolbox

PyTorch · CUDA / C++ · OpenCV · MONAI · Python · ONNX · Docker · Git

Reach me

Email · LinkedIn · Portfolio

Pinned Loading

  1. 3D-Image-Matching-with-LoFTR 3D-Image-Matching-with-LoFTR Public

    Jupyter Notebook 1

  2. pytorch/pytorch pytorch/pytorch Public

    Tensors and Dynamic neural networks in Python with strong GPU acceleration

    Python 102k 28.7k

  3. Project-MONAI/MONAI Project-MONAI/MONAI Public

    AI Toolkit for Healthcare Imaging

    Python 8.6k 1.6k

  4. opencv/opencv-python opencv/opencv-python Public

    Automated CI toolchain to produce precompiled opencv-python, opencv-python-headless, opencv-contrib-python and opencv-contrib-python-headless packages.

    Python 5.3k 1k

  5. rmaphoh/RETFound rmaphoh/RETFound Public

    Vision Foundation Models for Medical AI, including RETFound, DINOv2, DINOv3

    Python 663 159

  6. MCG-NJU/SAM2-Plus MCG-NJU/SAM2-Plus Public

    SAM 2++: Tracking Anything at Any Granularity

    Python 70 5