CellVision
Academic / Computer Vision
A blood-cell analysis platform spanning segmentation, labeling, active learning, correction, and reporting.
Overview
CellVision combines several steps of blood-cell image analysis in one Streamlit-based workflow.
The problem
Segmentation, labeling, correction, model evaluation, and reporting need to remain connected so human feedback can improve classifier quality.
My contribution
Built the end-to-end workflow, added human-in-the-loop labeling and active learning, and designed result and dashboard views.
System or workflow
- 01Segment cells from source images.
- 02Label and review extracted samples.
- 03Train and refine the classifier with active learning.
- 04Correct predictions and inspect final reporting views.
Technical highlights
- 01End-to-end workflow from image segmentation to prediction review
- 02Human-in-the-loop labeling and active learning
- 03Result and dashboard views for model evaluation and cell distribution