First iteration of MicroLens · Apr 24 · 24k VQA across 9 mixed microscale categories (pollen, algae, yeast, minerals, snow, tardigrades, plant disease, PCB, zooplankton) · proof-of-concept · superseded by v2/v3 · research only, not a medical device

MicroLens v1

Research model · Apache 2.0 · Not a medical device · Not a certified instrument · Use at your own risk. Outputs are statistical pattern matches against training data, not analytical measurements. See full disclaimer below.

The first iteration of the MicroLens series. Trained April 24, 2026 on 24,325 image-question-answer pairs across nine mixed microscale categories: pollen, algae, yeast, minerals, snowflakes, tardigrades, plant disease, PCB defects, zooplankton.

ollama run brinzaengineeringai/microlens-v1

Why this model is here

This is a historical artefact, kept on the registry to document the project’s progression. v1 was a proof-of-concept run on a deliberately broad training set — the goal was to verify that Gemma 4 E2B could be fine-tuned for microscopy at all on consumer hardware. It worked, with a final training loss of 0.42 after one epoch (about three hours on an RTX 3090 Ti).

For real production work, use v2 (terse parser-friendly output) or v3 (rich four-section descriptions). Both are trained on a much larger and deeper microscopy-only dataset (122,399 VQA pairs, 146 genera, 3 epochs).

What v1 covers

Nine categories with roughly equal representation:

Pollen grains (POLEN23E)
Algae cells (high-throughput detection dataset)
Yeast cells (TU Darmstadt, EMBC 2023)
Mineral specimens
Snowflakes (MASCDB and Colorado collections)
Tardigrades (Wikimedia Commons + EOL)
Plant disease (PlantDoc)
PCB inspection (KubeEdge IANVS)
Zooplankton (ZooScan crustacea)

Full source attribution and licensing in the docs/LICENSE_AUDIT.md of the repo.

Sample output

v1 produces descriptive text rather than the rigid one-line format of v2:

The image shows a tardigrade. Distinctive four-legged segmented body, visible digestive tract, translucent yellowish hue.

Pyrite with iridescent surface, likely due to thin-film interference or surface oxidation.

When to pick v1 over v2/v3

Honestly, rarely. Pick v1 if:

You need broader subject coverage (minerals, snowflakes, PCB defects, etc) that v2/v3 don’t cover
You want to compare iterations and study how data size and category focus affect a small VLM

For specific microscopy taxonomy (diatoms, plankton, fungal spores), v2/v3 are stronger.

Intended use and full disclaimer

MicroLens v1 is a research and educational artefact published under Apache 2.0. It is a fine-tuned neural network, not a regulated instrument.

Designed for:

Citizen-science screening
Taxonomy teaching and student labs
ML research, dataset benchmarking, model comparison
Pre-classification stages of professional pipelines, where every result is verified by a qualified person before any decision is made

This model is NOT, and must not be treated as:

A medical device, in-vitro diagnostic (IVD), or clinical decision-support tool
A regulatory-compliant water-quality measurement instrument (no ISO 17025, EPA, EU WFD, or equivalent certification)
A substitute for a trained taxonomist or accredited laboratory analysis
A calibrated, validated, or peer-reviewed analytical method

The model’s output is a probabilistic pattern match against the training data distribution, not a physical or analytical measurement. The model can be confidently wrong, particularly on subjects outside its training distribution.

No warranty. This software is provided “AS IS”, without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and non-infringement. In no event shall the author or contributors be liable for any claim, damages or other liability, whether in an action of contract, tort, or otherwise, arising from, out of, or in connection with the model or the use or other dealings in the model.

You assume all risk when downloading, deploying, modifying, or using this model on your own hardware. Always have qualified personnel verify any result that informs a regulatory, environmental, clinical, or health-related decision.

Built with

Unsloth — FastVisionModel with 4-bit QLoRA, LoRA rank 16, learning rate 2e-4
llama.cpp + mtmd — multimodal inference at the edge
Gemma 4 E2B-it — base model, Apache 2.0