Remote · Upwork Freelancer · Scientific Imaging Engineer
Python · OpenCV · PyTorch · Large-Scale Image Processing
I design and ship production-grade image processing pipelines for scientific microscopy — from raw microscope tiles to quantifiable results.
Clients come to me when they have large imaging datasets (gigabytes to terabytes per experiment) and need:
| Problem | What I Deliver |
|---|---|
| Stitch 10,000+ FOV tiles into one seamless whole-slide image | Memory-efficient pyramidal TIFF + libvips pipeline |
| Automatically find, count, and segment cells in microscopy images | YOLO / U-Net / Watershed with ONNX inference |
| Decode fluorescence barcodes across 1020 hybridization cycles | Multi-channel registration + Cy3/Cy5 ratio classification |
| Align multi-camera arrays with sub-pixel accuracy | Chessboard calibration + homography estimation |
| Run heavy image jobs on HPC clusters | SGE/Slurm schedulers, multiprocessing, error recovery |
BioChip huge-chip pipeline — 117,800 barcodes per cycle, 18 cycles
| Repo | BMK_huge_decode_program_S5000 |
| Stack | Python, OpenCV, YOLOv5 (ONNX), libvips, OpenSlide, NumPy, SciPy |
| What it does | FOV stitching, chip ROI search, barcode center detection, geometric correction, Cy3/Cy5 signal decoding, HPC-cluster orchestration |
| Result | Fully automated decoding for 380x310 arrays; handles 40 GB+ whole-slide images |
3x3 camera array system with interactive shift-and-sum refocusing
| Repo | Light-field |
| Stack | MATLAB, OpenCV, ORB Features, RANSAC Homography, GUI |
| What it does | Chessboard calibration across 9 cameras, perspective rectification, shift-and-sum refocus, interactive MATLAB GUI |
| Result | Real-time focal-plane adjustment from a single 9-camera exposure |
- Biomarker-cell_seg — Whole-slide cell boundary extraction, ROI analysis, spatial barcode-to-cell mapping
- Python, OpenCV, ONNX Runtime, libvips, scikit-image
- BMK-Huge-img-program — Register tiled FOVs, pyramidal TIFF composition, O(N log N) for 10K+ tiles
- Biomarker-BMChiper-no-div — Whole-slide alignment, array localization, stitching
- Python, OpenCV, ORB/SIFT, RANSAC Homography, libvips
- Decode_20X — Cross-cycle FOV registration + Cy3/Cy5 ratio-based base classification
- Python, OpenCV, libvips, NumPy
- ZTE2020Fourier-Preliminary — FFT experiments, spectrum analysis, frequency-domain filtering
- Python, NumPy FFT, Matplotlib
Python ecosystem
NumPy, SciPy, Pandas — numerical / signal / data
OpenCV, scikit-image, tifffile — computer vision & imaging
PyTorch, scikit-learn, ONNX Runtime — ML & inference
libvips, OpenSlide, h5py — large-image I/O
Infrastructure
Linux, Bash, Docker, Git
SGE / Slurm — HPC job schedulers
CUDA — GPU acceleration (PyTorch, OpenCV dnn)
Image formats: TIFF, BigTIFF, OME-TIFF, MRXS (OpenSlide), HDF5
- Whole-slide / large-image stitching and registration
- Cell segmentation and microscopy image analysis
- Fluorescence / spatial transcriptomics decoding pipelines
- Camera calibration and geometric transformation
- Custom computer vision for scientific instruments
- HPC cluster pipelines (SGE, Slurm, multiprocessing)
- One-off script to production pipeline conversion
| GitHub | github.com/ofen1996 |
| 526083628@qq.com | |
| Location | Remote |
Open to Upwork contracts and technical collaborations. Reply to this repo or drop an email — I typically respond within 24 hours.