envision-classifier 0.1.0__tar.gz

envision_classifier-0.1.0/LICENSE.md

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 FAIR Data Innovations Hub
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE 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 NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS 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 SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

envision_classifier-0.1.0/PKG-INFO

@@ -0,0 +1,120 @@
+Metadata-Version: 2.4
+Name: envision-classifier
+Version: 0.1.0
+Summary: Few-shot classifier for detecting eye imaging datasets
+License: MIT
+License-File: LICENSE.md
+Keywords: eye imaging,ophthalmology,OCT,fundus,retina,machine learning,dataset discovery,classification,setfit,few-shot,fair-data
+Author: FAIR Data Innovations Hub
+Author-email: contact@fairdataihub.org
+Requires-Python: >=3.10,<4.0
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Natural Language :: English
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering :: Image Processing
+Classifier: Topic :: Scientific/Engineering :: Medical Science Apps.
+Requires-Dist: click (>=8.0,<9.0)
+Requires-Dist: datasets (>=2.14.0)
+Requires-Dist: huggingface-hub (>=0.20.0)
+Requires-Dist: scikit-learn (>=1.3.0)
+Requires-Dist: setfit (>=1.0.0)
+Requires-Dist: torch (>=2.0.0)
+Requires-Dist: transformers (>=4.35.0)
+Project-URL: Documentation, https://envision-classifier.readthedocs.io
+Project-URL: Homepage, https://github.com/EyeACT/envision-classifier
+Project-URL: Repository, https://github.com/EyeACT/envision-classifier
+Description-Content-Type: text/markdown
+
+# envision-classifier
+
+SetFit few-shot classifier for identifying eye imaging datasets from scientific metadata.
+
+Part of the [EyeACT](https://github.com/EyeACT) project by the [FAIR Data Innovations Hub](https://fairdataihub.org).
+
+## Installation
+
+```bash
+pip install git+https://github.com/EyeACT/envision-classifier.git
+```
+
+## Python API
+
+```python
+from envision_classifier import EyeImagingClassifier
+
+# Downloads model from HuggingFace on first use
+clf = EyeImagingClassifier()
+
+# Classify a single record
+result = clf.classify("Retinal OCT dataset for diabetic retinopathy")
+print(result)
+# {'label': 'EYE_IMAGING', 'confidence': 0.999, 'probabilities': {...}}
+
+# Classify a batch
+results = clf.classify_batch([
+    "Retinal fundus photography dataset for glaucoma screening",
+    "COVID-19 genome sequencing data",
+    {"title": "OCT images", "description": "Macular degeneration scans"},
+])
+
+# Use a local model instead of downloading
+clf = EyeImagingClassifier(model_path="./my_model")
+```
+
+## CLI
+
+After installing, the `envision-classifier` command is available:
+
+```bash
+# Classify a text string
+envision-classifier classify --text "Retinal OCT dataset for diabetic retinopathy"
+
+# Classify from a JSON file
+envision-classifier classify records.json
+
+# Pipe JSON via stdin
+echo '{"title": "Fundus images", "description": "DR screening"}' | envision-classifier classify
+
+# Train a new model from built-in training data
+envision-classifier train --output ./my_model
+
+# Show model info and training data counts
+envision-classifier info
+```
+
+## Classification Labels
+
+| Label | Description |
+|-------|-------------|
+| **EYE_IMAGING** | Actual eye imaging datasets (fundus, OCT, OCTA, cornea) |
+| **EYE_SOFTWARE** | Code, tools, models for eye imaging (no actual data) |
+| **EDGE_CASE** | Eye research papers, reviews, non-imaging data |
+| **NEGATIVE** | Not eye-related |
+
+## Model
+
+- **Base model**: `sentence-transformers/all-mpnet-base-v2` (768-dim)
+- **Training data**: 474 curated examples (77 EYE_IMAGING, 48 EYE_SOFTWARE, 79 EDGE_CASE, 270 NEGATIVE)
+- **Test accuracy**: 0.937, **macro F1**: 0.902
+- **Spot-check**: 29/33 (87.9%)
+- **Model weights**: [fairdataihub/envision-eye-imaging-classifier](https://huggingface.co/fairdataihub/envision-eye-imaging-classifier)
+
+## Related
+
+- [envision-discovery](https://github.com/EyeACT/envision-discovery) -- Full pipeline (scraping + classification + export)
+- [Model on HuggingFace](https://huggingface.co/fairdataihub/envision-eye-imaging-classifier)
+
+## License
+
+MIT
+

envision_classifier-0.1.0/README.md

@@ -0,0 +1,82 @@
+# envision-classifier
+
+SetFit few-shot classifier for identifying eye imaging datasets from scientific metadata.
+
+Part of the [EyeACT](https://github.com/EyeACT) project by the [FAIR Data Innovations Hub](https://fairdataihub.org).
+
+## Installation
+
+```bash
+pip install git+https://github.com/EyeACT/envision-classifier.git
+```
+
+## Python API
+
+```python
+from envision_classifier import EyeImagingClassifier
+
+# Downloads model from HuggingFace on first use
+clf = EyeImagingClassifier()
+
+# Classify a single record
+result = clf.classify("Retinal OCT dataset for diabetic retinopathy")
+print(result)
+# {'label': 'EYE_IMAGING', 'confidence': 0.999, 'probabilities': {...}}
+
+# Classify a batch
+results = clf.classify_batch([
+    "Retinal fundus photography dataset for glaucoma screening",
+    "COVID-19 genome sequencing data",
+    {"title": "OCT images", "description": "Macular degeneration scans"},
+])
+
+# Use a local model instead of downloading
+clf = EyeImagingClassifier(model_path="./my_model")
+```
+
+## CLI
+
+After installing, the `envision-classifier` command is available:
+
+```bash
+# Classify a text string
+envision-classifier classify --text "Retinal OCT dataset for diabetic retinopathy"
+
+# Classify from a JSON file
+envision-classifier classify records.json
+
+# Pipe JSON via stdin
+echo '{"title": "Fundus images", "description": "DR screening"}' | envision-classifier classify
+
+# Train a new model from built-in training data
+envision-classifier train --output ./my_model
+
+# Show model info and training data counts
+envision-classifier info
+```
+
+## Classification Labels
+
+| Label | Description |
+|-------|-------------|
+| **EYE_IMAGING** | Actual eye imaging datasets (fundus, OCT, OCTA, cornea) |
+| **EYE_SOFTWARE** | Code, tools, models for eye imaging (no actual data) |
+| **EDGE_CASE** | Eye research papers, reviews, non-imaging data |
+| **NEGATIVE** | Not eye-related |
+
+## Model
+
+- **Base model**: `sentence-transformers/all-mpnet-base-v2` (768-dim)
+- **Training data**: 474 curated examples (77 EYE_IMAGING, 48 EYE_SOFTWARE, 79 EDGE_CASE, 270 NEGATIVE)
+- **Test accuracy**: 0.937, **macro F1**: 0.902
+- **Spot-check**: 29/33 (87.9%)
+- **Model weights**: [fairdataihub/envision-eye-imaging-classifier](https://huggingface.co/fairdataihub/envision-eye-imaging-classifier)
+
+## Related
+
+- [envision-discovery](https://github.com/EyeACT/envision-discovery) -- Full pipeline (scraping + classification + export)
+- [Model on HuggingFace](https://huggingface.co/fairdataihub/envision-eye-imaging-classifier)
+
+## License
+
+MIT

envision_classifier-0.1.0/envision_classifier/__init__.py

@@ -0,0 +1,25 @@
+"""
+envision-classifier: Eye Imaging Dataset Classifier
+
+A 4-class SetFit classifier for detecting eye imaging datasets:
+  - EYE_IMAGING: Actual eye imaging datasets (fundus, OCT, OCTA, etc.)
+  - EYE_SOFTWARE: Code, models, tools for eye imaging
+  - EDGE_CASE: Eye research papers, reviews, borderline items
+  - NEGATIVE: Unrelated domains
+
+Usage:
+    >>> from envision_classifier import EyeImagingClassifier
+    >>> clf = EyeImagingClassifier()
+    >>> clf.classify("Retinal OCT dataset for diabetic retinopathy")
+    {'label': 'EYE_IMAGING', 'confidence': 0.999, 'probabilities': {...}}
+"""
+
+__version__ = "0.1.0"
+__author__ = "James O'Neill"
+
+from .classifier import EyeImagingClassifier, LABELS
+
+__all__ = [
+    "EyeImagingClassifier",
+    "LABELS",
+]

envision_classifier-0.1.0/envision_classifier/__main__.py

@@ -0,0 +1,4 @@
+"""Allow running as python -m envision_classifier."""
+from .cli import cli
+
+cli()

envision_classifier-0.1.0/envision_classifier/classifier.py

@@ -0,0 +1,813 @@
+"""
+ENVISION: Eye Imaging Dataset Classification
+
+SetFit Few-Shot Classifier for Eye Imaging Dataset Detection
+Uses sentence-transformers/all-mpnet-base-v2 sentence transformer with 4-class classification:
+  - 3: EYE_IMAGING - Actual eye imaging datasets (fundus, OCT, OCTA, cornea, etc.)
+  - 2: EYE_SOFTWARE - Code, tools, models for eye imaging (no actual data)
+  - 1: EDGE_CASE - Eye research (papers, reviews, non-imaging data)
+  - 0: NEGATIVE - Not eye-related at all
+"""
+
+import os
+import re
+import warnings
+from html import unescape
+from pathlib import Path
+
+import torch
+
+warnings.filterwarnings("ignore")
+
+# Model configuration
+BASE_MODEL_NAME = "sentence-transformers/all-mpnet-base-v2"
+HF_MODEL_REPO = "fairdataihub/envision-eye-imaging-classifier"
+LABELS = ["NEGATIVE", "EDGE_CASE", "EYE_SOFTWARE", "EYE_IMAGING"]
+
+# ============================================================
+# TRAINING DATA - Curated examples for few-shot learning
+# ============================================================
+
+# EYE_IMAGING (label=3): Actual eye imaging datasets with real image data
+# Cleaned: removed 21 misplaced examples (reviews, software, stats, metabolomics)
+EYE_IMAGING_EXAMPLES = [
+    "Dataset from fundus images for the study of diabetic retinopathy progression",
+    "Optical Coherence Tomography Angiography-OCTA Dataset for Diabetic Retinopathy",
+    "Indian Diabetic Retinopathy Image Dataset (IDRiD) — Segmentation and Grading Challenge",
+    "Retinal Fundus Multi-Disease Image Dataset (RFMiD) 2.0",
+    "Rotterdam EyePACS AIROGS train set - fundus images for glaucoma detection",
+    "JustRAIGS challenge training data set - Justified Referral in AI Glaucoma Screening",
+    "OLIVES Dataset: Ophthalmic Labels for Investigating Visual Eye Semantics",
+    "RVD: A Handheld Device-Based Fundus Video Dataset for Retinal Vessel Segmentation",
+    "High-resolution structural and functional retinal imaging in mice",
+    "Multi-modal spatiotemporal phenotyping of human retinal organoid development",
+    "2023 IEEE SPS Video and Image Processing (VIP) Cup: Ophthalmic Biomarker Detection",
+    "DERMA-OCTA: OCT Angiography images for skin microvascular analysis",
+    "Abca4 inhibition in a cone-rich rodent leads to Stargardt Disease OCT and fundus images",
+    "Diabetic Glaucoma dataset combining ORIGA, REFUGE, ACRIMA fundus images",
+    "OCT-A mosaicking dataset for retinal vessel analysis",
+    "UTHealth - Fundus and Synthetic OCT-A Dataset (UT-FSOCTA)",
+    "Myopic Maculopathy Analysis Challenge 2023 - fundus image dataset",
+    "Retinal vessel segmentation challenge dataset DRIVE STARE CHASE_DB1",
+    "Glaucoma OCT dataset with RNFL thickness measurements",
+    "Age-related macular degeneration OCT B-scan image dataset",
+    "Heidelberg Spectralis OCT scans for diabetic macular edema",
+    "Zeiss Cirrus HD-OCT dataset for glaucoma progression analysis",
+    "Topcon 3D OCT fundus and cross-sectional images",
+    "Fluorescein angiography dataset for choroidal neovascularization",
+    "Fundus photography dataset for optic disc and cup segmentation",
+    "Retinal OCT images with drusen and geographic atrophy labels",
+    "OCTA dataset showing foveal avascular zone measurements",
+    "Corneal topography images for keratoconus detection",
+    "Anterior segment OCT dataset for angle closure glaucoma",
+    "Slit lamp photography dataset for cataract grading",
+    "Multi-Disease Detection in Retinal Imaging dataset",
+    "Retinal Wave Dataset - calcium imaging of developing retina",
+    "Evaluation benchmark for natural robustness of retinal vessel segmentation",
+    "Probabilistic volumetric speckle suppression in OCT using deep learning",
+    "Optical coherence tomography radiation cataract image dataset",
+    "HRF-Seg+: A Multi-Structure Annotated Fundus Image Dataset",
+    "Multimodal OCTA and Fundus Image dataset for diabetic retinopathy detection",
+    "Iraqi Retinal Fundus Diabetic Retinopathy Dataset IRFDRD",
+    "OCT Fundus Registration dataset for multimodal retinal analysis",
+    "Topological characterization of the retinal microvascular network",
+    "JRC-Multi-Modal Retinal Vessel Segmentation dataset",
+    "A Fundus Image Dataset for Domain Generalization in Joint Optic Disc and Cup Segmentation",
+    "CeraMIRScan: Mid-infrared OCT Scan Dataset for ophthalmic applications",
+    "qtOCT: quantitative transmission optical coherence tomography dataset",
+    "Mammalian animal and human retinal organ culture imaging data",
+    "Spontaneous retinal reperfusion of capillary nonperfusion OCT and fundus",
+    "Analysis on Multimodal Imaging of stealth Choroidal Neovascularization OCTA",
+    "Longitudinal changes in retinal microstructures OCT imaging data",
+    "Macular Drusen histology and OCT correlation dataset",
+    "Generalized Analysis of Vessels in Eye GAVE Challenge dataset",
+    "Automated fundus image quality assessment and segmentation dataset",
+    "Diabetic Retinopathy Detection using Retinal Images dataset",
+    "Optic disc localization using graph traversal algorithm dataset",
+    "An Image Processing Algorithm to Detect Exudates in Fundus Images",
+    "Binary operation based hard exudate detection fundus dataset",
+    "EXUDATES DETECTION FROM DIGITAL FUNDUS IMAGE dataset",
+    "COMPARATIVE STUDY OF DIABETIC RETINOPATHY K-NN dataset",
+    "AN AUTOMATIC SCREENING METHOD TO DETECT OPTIC DISC dataset",
+    "Fundus Fluorescein Angiography imaging dataset",
+    "Structural-Functional Transition in Glaucoma Assessment imaging data",
+    "Identification of ocular disease from fundus images using deep learning",
+    "To Assess Characteristics of Individuals with Disc Hemorrhage fundus imaging",
+    "Prevalence Risk Factors Clinical Correlates Age-related Macular Degeneration imaging",
+    "ResNet-n/DR Automated diagnosis of diabetic retinopathy fundus dataset",
+    "Eye fundus oxygenation mapping from color retinographs dataset",
+    "Machine learning classifiers for detection of glaucoma OCT dataset",
+    "Enhancing Retinal Disease Detection with Swin Transformer fundus dataset",
+    "Development of hybrid framework to characterize red lesions in fundus images",
+    "Diabetic retinopathy classification using deep convolutional neural networks fundus",
+    "EARLY DETECTION OF HIGH BLOOD PRESSURE AND DIABETIC RETINOPATHY fundus images",
+    "Diagnosis Of Diabetic Retinopathy: fundus image analysis dataset",
+    "FUNDUS IMAGES FOR DIAGNOSIS OF DIABETIC RETINOPATHY dataset",
+    "Retinal status analysis method based on feature extraction OCT dataset",
+    "Data from Inactivation of adenosine receptor retinal imaging",
+    "Polarisation camera dSTORM datasets of retinal cells",
+    "Scanning dynamic light scattering optical coherence tomography retinal flow",
+    "MedIMeta: multi-domain medical imaging including retinal fundus",
+]
+
+# EYE_SOFTWARE (label=2): Code, tools, models for eye imaging (NOT actual data)
+# Added: misplaced software from EYE_IMAGING + EDGE_CASE, spot-check examples
+EYE_SOFTWARE_EXAMPLES = [
+    "linchundan88/Fundus-image-preprocessing: fundus image preprocessing Python code",
+    "NIH-NEI/oct-image-segmentation-models: v0.8.2 trained model weights",
+    "optic-nerve-cnn: First version of the software neural network",
+    "Corneal-Endothelium-Data-Annotation-Tool: annotation labeling software",
+    "FundusImageToolbox: Python package fundus image processing library",
+    "OCTAVA: open-source toolbox quantitative analysis OCT angiography",
+    "RetinoNet-DR-Classification: deep learning code diabetic retinopathy",
+    "oct-to-tiff: command line tool OCT angiography converter",
+    "optic-disc-segmentation-drishtigs: segmentation algorithm implementation",
+    "QiYanPitt/AMDprogressCNN: Late AMD Fundus Image Prediction model",
+    "Deep learning model weights trained on fundus images PyTorch",
+    "Pretrained neural network OCT segmentation model weights only",
+    "ResNet-50 classifiers trained on retinal fundus model weights",
+    "ONNX model weights retinal vessel segmentation inference",
+    "PyTorch checkpoint OCT layer segmentation neural network",
+    "Segmentation model weights diabetic retinopathy detection",
+    "Python package retinal image preprocessing pip install",
+    "MATLAB toolbox fundus image analysis code only",
+    "ImageJ plugin OCT visualization and measurement",
+    "Fiji macro optic nerve fiber layer analysis",
+    "R package macular thickness statistical analysis",
+    "Source code implementation retinal vessel extraction",
+    "Algorithm implementation optic disc detection CNN",
+    "Code repository deep learning diabetic retinopathy",
+    "GitHub release fundus segmentation neural network",
+    "Jupyter notebook tutorial OCT image classification",
+    "Caserel: Open Source Software Computer-aided Segmentation Retinal Layers",
+    "duke-lungmap-team/odifmap: image processing code publication",
+    "young-oct/OCT-denoising: denoising algorithm code repository",
+    "costapt/vess2ret: vessel to retina synthesis code",
+    # Moved from EYE_IMAGING (software/tools, not datasets)
+    "EyeLab: Python package for OCT and fundus image processing",
+    "Fiji-mCNVImageAnalysisTool for choroidal neovascularization in OCTA",
+    "Automatic Choroid Vascularity Index Calculation in OCT Images",
+    "ResNet-50 classifiers and diffusion models trained on retinal fundus images",
+    "AMikroulis/octopus OCT image processing dataset",
+    "anithaj17/RetinoNet-DR-Classification fundus image dataset",
+    # Moved from EDGE_CASE (clearly software/tools)
+    "Python package for retinal image preprocessing",
+    "Deep learning framework for fundus image segmentation code only",
+    "OCT image reconstruction algorithm implementation",
+    "Retinal vessel extraction software repository",
+    "Optic disc detection neural network model weights",
+    "Diabetic retinopathy grading API documentation",
+    "Fundus image augmentation library code",
+    "DICOM viewer for ophthalmic images software",
+    "OCT visualization toolkit implementation",
+    "Retinal layer segmentation algorithm code repository",
+    # Spot-check derived (model weights / recording device code)
+    "ANNs pre-trained on Retinal Waves model weights",
+    "Flexible corneal neurotechnology reveals in-vivo pathological retinal oscillations recording device",
+]
+
+# EDGE_CASE (label=1): Eye/vision research but NOT actual imaging datasets
+# Cleaned: removed misplaced software→EYE_SOFTWARE, non-eye→NEGATIVE; added eye metabolomics
+EDGE_CASE_EXAMPLES = [
+    "A Review of Deep Learning Methods for Diabetic Retinopathy Detection",
+    "Survey of Machine Learning Techniques for Glaucoma Diagnosis",
+    "Advances in Optical Coherence Tomography Technology Review Article",
+    "Clinical Guidelines for Diabetic Eye Screening",
+    "Comparison of OCT Devices: A Systematic Review",
+    "Deep Learning in Ophthalmology: A Comprehensive Review",
+    "Artificial Intelligence in Retinal Disease Detection Review",
+    "State of the Art in Fundus Image Analysis Survey",
+    "Future Directions in Ophthalmic Imaging Technology",
+    "Machine Learning for Age-Related Macular Degeneration: A Review",
+    "Genetic factors in age-related macular degeneration GWAS meta-analysis",
+    "Molecular mechanisms of retinal ganglion cell death in glaucoma",
+    "Pharmacological treatment options for diabetic macular edema",
+    "Risk factors for progression of diabetic retinopathy clinical study",
+    "Visual acuity outcomes after anti-VEGF therapy clinical trial",
+    "Intraocular pressure measurement techniques comparison study",
+    "Epidemiology of myopia in Asian populations survey",
+    "Cataract surgery outcomes in diabetic patients retrospective analysis",
+    "Color vision deficiency prevalence in school children",
+    "Visual field testing protocols for glaucoma clinical practice",
+    "Electronic health records analysis of glaucoma treatment patterns",
+    "Patient-reported outcomes in dry eye disease questionnaire data",
+    "Healthcare costs of diabetic eye disease economic analysis",
+    "Ophthalmologist workforce distribution geographic study",
+    "Barriers to diabetic eye screening qualitative interview data",
+    "Adherence to glaucoma medication patient diary data",
+    "Visual impairment and quality of life survey responses",
+    "Telemedicine in ophthalmology implementation analysis",
+    "Eye care access in rural communities demographic data",
+    "Waiting times for cataract surgery administrative data",
+    # Eye tracking / gaze (eye-related but not imaging datasets)
+    "Eye tracking data for attention research",
+    "Gaze estimation dataset for human-computer interaction",
+    "Pupil dilation response to emotional stimuli",
+    "Saccade patterns in reading comprehension study",
+    "Visual search behavior eye movement data",
+    "Fixation duration analysis for cognitive load",
+    "Eye blink detection for drowsiness monitoring",
+    "Iris recognition biometric dataset",
+    "Facial expression analysis including eye region",
+    "Driver attention monitoring eye tracking",
+    # Animal eye development (eye-related but not human imaging)
+    "Drosophila compound eye development gene expression",
+    "Zebrafish eye regeneration molecular analysis",
+    "Mouse retinal development transcriptomics",
+    "Chicken embryo eye formation RNA sequencing",
+    "Frog photoreceptor electrophysiology recordings",
+    "Squid giant axon eye homolog studies",
+    "Insect compound eye optics physics modeling",
+    "Cephalopod camera eye evolution genomics",
+    "Spider eye arrangement morphological analysis",
+    "Mantis shrimp visual system spectral analysis",
+    # Other eye measurements (not imaging)
+    "Digital fundus thermometry for fever screening",
+    "Ocular surface temperature measurement",
+    "Tear film stability analysis without imaging",
+    "Contrast sensitivity function psychophysics",
+    "Dark adaptation curve measurements",
+    "Electroretinography signal analysis only",
+    "Visual evoked potential recordings",
+    "Optical properties of crystalline lens in vitro",
+    "Corneal biomechanics simulation data",
+    "Aqueous humor proteomics analysis",
+    # Moved from EYE_IMAGING (eye-related but not imaging datasets)
+    "GWAS Summary Statistics For Eye Imaging Traits",
+    "Fundus vessel phenotypes quantitative trait dataset",
+    "Retinal S-cone specific anatomical and physiological data",
+    "Nonlinear spatial integration allows the retina to detect the direction of motion",
+    "Thrombospondin-1 Mediates Axon Regeneration in Retinal Ganglion Cells",
+    "Optic nerve injury impairs intrinsic mechanisms underlying early eye imaging",
+    "Circuit mechanisms underlying embryonic retinal waves dataset",
+    "Data from Analysis of potential ischemic effect of intravitreal anti-VEGF OCT",
+    "Photodynamic Ocular Drug Delivery System with OCT monitoring",
+    "Data to Choroidal changes in intermediate age-related macular degeneration",
+    # Spot-check derived (eye-related but not imaging)
+    "Perspectives and Limitations of Mesenchymal Stem Cell-Based Therapy for Corneal injuries",
+    "Mesoporous Silica Nanocarriers of siRNA for Retinal Delivery drug mechanism",
+    # Eye metabolomics (eye tissue but metabolomics, not imaging)
+    "Metabolomics of mouse retina and optic nerve",
+    "Metabolomics of ocular hypertensive rat optic nerve",
+    "NMR spectroscopy-based metabolomics of organotypic retinal explants",
+    "Vitreous humor metabolomics profiling in diabetic retinopathy and retinal detachment",
+    "Spatial metabolomics of primate retina macula and periphery regions",
+    "Aqueous humor metabolome analysis in glaucoma patients",
+    "Lipidomics of retinal pigment epithelium in age-related macular degeneration",
+]
+
+# NEGATIVE (label=0): Clearly not eye-related
+# Added: non-eye medical imaging from EDGE_CASE, spot-check confounders
+NEGATIVE_EXAMPLES = [
+    "Climate change impact on coral reef ecosystems dataset",
+    "COVID-19 genome sequencing and variant analysis",
+    "Electric vehicle battery performance testing data",
+    "Social media sentiment analysis Twitter dataset",
+    "Stock market prediction historical price data",
+    "Natural language processing benchmark dataset",
+    "Robot navigation and path planning simulation",
+    "Music genre classification audio features",
+    "Speech recognition multilingual corpus",
+    "Protein structure prediction AlphaFold data",
+    "Urban traffic flow optimization dataset",
+    "Earthquake seismic wave recordings",
+    "Satellite imagery land use classification",
+    "Agricultural crop yield prediction dataset",
+    "Air quality monitoring sensor data",
+    "Ocean temperature salinity measurements",
+    "Forest fire detection and spread modeling",
+    "Wind turbine power output dataset",
+    "Solar panel efficiency measurements",
+    "Smart grid energy consumption patterns",
+    "Human gut microbiome metagenomic sequencing",
+    "Cancer cell line drug response screening",
+    "Plant root architecture phenotyping images",
+    "Bacterial biofilm formation time lapse",
+    "Yeast protein interaction network",
+    "Mouse brain connectome neural tracing",
+    "Human genome whole exome sequencing",
+    "Single cell RNA sequencing pancreas",
+    "Epigenome methylation profiling data",
+    "Metabolomics of liver disease samples",
+    "Image classification benchmark ImageNet",
+    "Object detection COCO dataset",
+    "Face recognition LFW dataset",
+    "Handwriting recognition MNIST digits",
+    "Autonomous driving perception dataset",
+    "Video action recognition UCF101",
+    "3D point cloud semantic segmentation",
+    "Document layout analysis dataset",
+    "Scene text recognition benchmark",
+    "Pose estimation human keypoints",
+    "Compiler optimization benchmark suite",
+    "Database query performance testing",
+    "Network intrusion detection logs",
+    "Software bug report classification",
+    "Code review comment sentiment",
+    "API usage pattern analysis",
+    "Container orchestration metrics",
+    "Microservice latency measurements",
+    "Cloud resource utilization data",
+    "DevOps pipeline performance metrics",
+    "Historical newspaper digitization project",
+    "Archaeological site survey mapping",
+    "Linguistic corpus for dialect analysis",
+    "Museum artifact catalog metadata",
+    "Legal case document classification",
+    "Political speech transcript analysis",
+    "Immigration policy document corpus",
+    "Educational assessment score data",
+    "Survey responses on housing affordability",
+    "Census demographic statistics",
+    "Weather forecast model output data",
+    "Cryptocurrency transaction network",
+    "Hotel review sentiment dataset",
+    "Recipe ingredient network analysis",
+    "Movie recommendation collaborative filtering",
+    "Book summary text generation",
+    "News article topic classification",
+    "Sports statistics player performance",
+    "FIGURES 1-10 in Taxonomic revision of genus species description",
+    "Figs 12-19 in Review of insect family Hemiptera Pentatomidae",
+    "FIGURES 45-53 in Introduction to Scydmaeninae Coleoptera beetles",
+    "FIGURE 6 in Additions to the description of new beetle species",
+    "Figs 7-11 in Review of Parachinavia insect taxonomy",
+    "FIGURES 14-19 in World genera of arthropod taxonomy review",
+    "FIGURE 15 in Combining morphological and molecular data new species",
+    "Figure 4 in The neurocranium of fish species morphology anatomy",
+    "FIGURES 64-68 in Franz and Nogunius genus description taxonomy",
+    "FIGURES 211-215 in Introduction to beetle family Coleoptera",
+    "Figs 1-5 in New species description and taxonomic placement",
+    "FIGURES 102-104 in curse of Horaeomorphus taxonomy revision",
+    "FIGURE 33 in Kirkegaardia polychaete worm new species",
+    "lymph node ultrasound image dataset pathology",
+    "PDAC tumour and vessel segmentation pancreatic cancer",
+    "Aortic valve calcification CT scan imaging",
+    "Atherosclerotic plaque OCT cardiovascular imaging",
+    "Lung nodule detection chest X-ray dataset",
+    "Brain MRI Alzheimer disease classification",
+    "Cardiac ultrasound echocardiography dataset",
+    "Mammography breast cancer detection images",
+    "Skin lesion dermoscopy melanoma dataset",
+    "Liver CT segmentation dataset",
+    "Fosil bivalvo fossil bivalve specimen",
+    "Fossil shell morphology museum specimen",
+    "Paleontology specimen 3D scan dataset",
+    "Hand-eye camera calibration robotics dataset",
+    "Robot eye camera sensor data manipulation",
+    "Machine vision inspection camera system",
+    "NAH rectangular plate Nearfield Acoustic Holography",
+    "Ultrasound transducer beam pattern dataset",
+    "Sonar imaging underwater acoustic data",
+    "Gaming leaderboard historical data",
+    "E-commerce product catalog data",
+    "Tourism destination visitor statistics",
+    "Fashion image style classification",
+    "Food image recognition dataset",
+    "Indoor scene recognition benchmark",
+    "Texture classification material images",
+    "Furniture detection room layout",
+    "Vehicle make model classification",
+    "Bird species identification dataset",
+    "Flower recognition 102 categories",
+    "Dog breed classification Stanford Dogs",
+    "Butterfly species identification images",
+    "Insect pest detection agricultural",
+    "Fish species classification underwater",
+    "Wildlife camera trap image dataset",
+    "Plankton microscopy classification",
+    "Cell microscopy segmentation HeLa",
+    "Pollen grain identification dataset",
+    "Mineral classification geological samples",
+    "Timber species identification wood",
+    "Fabric defect detection textile",
+    "Broadband acousto-optic modulators on Silicon Nitride photonics",
+    "Artifacts in Optical Projection Tomography general imaging",
+    "Optic flow and odometry data from intelrealsense camera robotics",
+    "Interstitial null-distance time-domain diffuse optical spectroscopy",
+    "iris-esmf-regrid Earth System Modeling Framework climate software",
+    "Altotiberina Low-angle normal fault seismic seismology",
+    "Refined Terrace Extraction Method geography terrain analysis",
+    "Lithospheric structure geological analysis dataset",
+    "integrated multi-scale approach to habitat modelling ecology",
+    "global variations in directional solar radiation exposure geography",
+    "IRIS Carbon Mapping Project carbon emissions dataset",
+    "Improved River Slope Datasets United States Hydrofabrics hydrology",
+    "Tracking Carboplatin Chemoresistance in Ovarian Cancer dataset",
+    "Subtype identification clear cell renal cell carcinoma kidney cancer",
+    "aortic dataset cardiovascular vessel segmentation",
+    "Deep learning aneurysm detection CT angiography brain vessels",
+    "Images from carotid artery patients cardiovascular disease",
+    "post-dive precordial subclavian Doppler ultrasound diving medicine",
+    "Deep Learning Segmentation Atherosclerotic Plaque cardiovascular",
+    "pLGG Radioimmunomics pediatric low-grade glioma brain tumor",
+    "Ex Vivo MRI Frontotemporal Lobar Degeneration brain imaging",
+    "Intracranial Sonodynamic Therapy brain treatment dataset",
+    "Adult female Aedes albopictus mosquito specimen imaging",
+    "Female pupa Aedes albopictus mosquito developmental imaging",
+    "Comparative larval ontogeny fish species developmental anatomy",
+    "Hadzinia ferrani Opiliones Nemastomatidae spider taxonomy",
+    "First record genus Tanaostigma Hymenoptera Chalcidoidea wasp taxonomy",
+    "Refractive index tomography chitin bristles chaetae marine worms",
+    "Methodology labeled image datasets entomological specimens insects",
+    "Newman-planar-elasticity computational physics simulation",
+    "Enhanced Photoactivity Carbon Nanodots Zinc Phthalocyanine photochemistry",
+    "Fluorescein-switching lateral flow assay chemistry biosensor",
+    "CtBP2 MD trajectories molecular dynamics protein simulation",
+    "All-atom accelerated molecular dynamics Filamin-A protein",
+    "QuantumScents molecular chemistry scent compound dataset",
+    "Influence Firing Temperature Silver-Aluminium Paste solar cell fabrication",
+    "Boron-Emitter Development TOPCon c-Si Solar Cells photovoltaics",
+    "Photo-physical characterization brominated fluorophore chemistry",
+    "MicroED datasets hemin biotin electron crystallography",
+    "Quantifying impact electric field computational physics",
+    "Accurate Modeling Bromide Iodide Hydration molecular chemistry",
+    "Carrier Diffusion Recombination semiconductor physics perovskite",
+    "Nanoparticle doping molten-core fiber optics materials",
+    "Grain orientation angle incidence beam polarization materials",
+    "Observing impacts luminescence spectroscopy materials",
+    "magnetic topology neutral trapping plasma physics tokamak",
+    "Unraveling hierarchical structure saturated monoacid triglycerides lipids",
+    "Training data bead stacks Zeiss microscope calibration beads",
+    "COLMAP outputs Gaussian Splatting Reconstruction 3D computer vision",
+    "Direct STORM imaging transcription element microscopy super-resolution",
+    "Cross-polarized light microscopy Coccospheres marine microfossils",
+    "Cryo-electron microscopy thin vitreous biological samples cryoEM",
+    "Raw confocal imaging FRAP protein dynamics general microscopy",
+    "Evaluation strategy image acquisition protocols confocal microscopy",
+    "STORM Vectashield datasets Tubulin cytoskeleton microscopy",
+    "Objective evaluation image quality planning CT radiation therapy",
+    "In Situ Volumetric Imaging FRESH 3D Bioprinted Constructs bioprinting",
+    "Thermal-plex fluidic-free rapid sequential multiplexed imaging proteomics",
+    "Histological validation per-bundle water diffusion brain tractography",
+    "Large-scale in vivo acquisition brain vasculature cerebral vessels",
+    "Chronic social defeat stress meningeal neutrophilia brain inflammation",
+    "Tracing pathways high-resolution tractography brain connectivity",
+    "BRAVE-NET Fully Automated Arterial Brain Vessel segmentation",
+    "Correlated variability primate superior colliculus brain neural",
+    "Widefield time-lapse Drosophila embryos developmental biology fly",
+    "Transcriptomic profiling immune cells pleural effusions lung cancer",
+    "3D-printed adapters standardized radiometric photometric calibration equipment",
+    "Patterns Gene Expression Splicing Allele-Specific Expression genomics",
+    "Modular Tunable Gene Expression Sensing synthetic biology circuits",
+    "FIGURE characterisation stem proliferating cells generic figure",
+    "Project Gap Junctions microelectrode array electrophysiology",
+    "Objective Autonomic Signatures Tinnitus hearing audiology",
+    "IML-DKFZ/fd-shifts machine learning code repository",
+    "Aspergillus flavus germination fungal pathogen imaging",
+    "Intrapartum Ultrasound Grand Challenge obstetric fetal imaging",
+    "Correlating Spectral Properties complex mineral samples mineralogy",
+    "YOLO Based Machine Learning general object detection computer vision",
+    "Freehand ultrasound without external trackers general ultrasound imaging",
+    "scRNAseq datasets cranial myogenic progenitors muscle development",
+    "Confocal fluorescence microscopy dentinal porosity dental imaging",
+    "Circadian rapid eye movement sleep expression sleep research polysomnography",
+    "Label-free metabolic fingerprinting motile mammalian spermatozoa fertility",
+    "OCT IMAGE DATASET RADIATION DERMATITIS skin dermatology",
+    "Tissue-Level Dimerization Analysis AtLEA proteins Arabidopsis plant biology",
+    "Ear Datasets hearing auditory speech recognition",
+    "Exchange interaction FAD biradical magnetic resonance chemistry",
+    "AF driver detection pulmonary vein area cardiac arrhythmia",
+    "Roman-Multi-Planetary-data astronomy exoplanet detection",
+    "Compatible interaction experiment Aegilops cylindrica wheat pathogen plant",
+    "Raw EOG Data electrooculography electrical eye movement recording",
+    "CEAP-360VR Continuous Physiological Behavioral Emotion VR annotation",
+    "Pultruded carbon fiber profiles 3D x-ray tomography composites materials",
+    "Optical electronic signal stabilization plasmonic fiber optic gas sensor",
+    "Propulsion nano microcones traveling ultrasound wave acoustic manipulation",
+    "Shear Shock Waves Haptic Holography Focused Ultrasound haptics",
+    "CeraMIRScan Mid-infrared OCT Ceramic Quality industrial inspection",
+    "Dosage effect Copy Number Variation Epilepsy genetic neurology",
+    "Field-Effect Transistor Plasmonic Fiber Optic Gate Electrode electronics",
+    "Early Onset TAAD cohort genetic cardiovascular aortic disease",
+    "Transcriptomic profiling of immune cells in pleural effusions identifies macrophages",
+    "fqjin/skin-segmentation skin lesion segmentation code",
+    "Subtype identification and clinical application of clear cell renal cell carcinoma",
+    "A dataset of global variations in directional solar radiation exposure for ocular surface",
+    "Pre-training with simulated ultrasound images for breast mass segmentation",
+    "aortic dataset for DB-SNet cardiovascular aortic segmentation",
+    "Ground truth labels for BRAVE-NET Fully Automated Arterial Brain Vessel Segmentation",
+    "AstroFatheddin/Roman-Multi-Planetary-data astronomy exoplanet",
+    "Dataset for Segmentation and Multi-Timepoint Tracking of 3D Cancer Organoids",
+    "Dataset_1 of AF driver detection in pulmonary vein area cardiac arrhythmia",
+    "Data from Dichoptic metacontrast masking functions to infer transmission delay",
+    "IRIS Carbon Mapping Project Curated Dataset carbon emissions",
+    # Moved from EDGE_CASE (non-eye medical imaging — clearly NEGATIVE)
+    "Brain MRI analysis for Alzheimer's disease detection",
+    "Cardiac CT angiography for coronary artery disease",
+    "Dermatology skin lesion classification dataset",
+    "Dental X-ray caries detection images",
+    "Chest X-ray pneumonia detection dataset",
+    "Mammography breast cancer screening images",
+    "Histopathology slide analysis for cancer diagnosis",
+    "Ultrasound imaging for liver disease assessment",
+    "PET scan analysis for neurological disorders",
+    "Spine MRI for degenerative disc disease",
+    # Moved from EDGE_CASE (non-eye OCT — clearly NEGATIVE)
+    "OCT for industrial material inspection dataset",
+    "Optical coherence tomography in dermatology skin imaging",
+    "OCT imaging of atherosclerotic plaque in arteries",
+    "Dental OCT for tooth structure analysis",
+    "OCT for art conservation painting analysis",
+    "Industrial OCT for semiconductor inspection",
+    "OCT in cardiology intravascular imaging",
+    "Non-destructive testing using OCT",
+    "OCT for pharmaceutical tablet coating analysis",
+    "Ceramic quality inspection using OCT",
+    # Spot-check confounders: brain/neuro segmentation (use "segmentation" vocabulary)
+    "ASHS-OAP atlas for automatic entorhinal cortex segmentation brain Alzheimer",
+    "FastSurferVINN Checkpoints brain segmentation model weights",
+    "Phospho-seq integrated multi-modal profiling intracellular protein dynamics brain organoid",
+    # Spot-check confounders: non-eye segmentation (use "segmentation" vocabulary)
+    "Doodleverse building and flood segmentation satellite imagery",
+    "Doodleverse water segmentation remote sensing",
+    "Beach sediment detection and segmentation coastal imagery",
+    "DTL-IceNet ice segmentation polar remote sensing",
+    "Vine Trunk Semantic Segmentation using Individual Vine Trunks agriculture",
+    # Spot-check confounders: plant OCT (use "OCT" vocabulary but not eye)
+    "Quantification of plant morphology and leaf thickness with OCT",
+    "Revealing real-time 3D in vivo pathogen dynamics in plants by label-free optical",
+    # Spot-check confounders: other misclassified records
+    "Knowledge-Guided ML can improve carbon cycle predictions in agriculture",
+    "PomerFish dataset for fishes across Pomerania freshwater waterbodies underwater",
+    "Integrating time-series analysis and deep learning to reconstruct RTS dynamics remote sensing",
+    # Spot-check: 4 additional missing NEGATIVE confounders
+    "Voronoi segmentation tutorial generic image processing",
+    "Audio-visual segmentation multimodal learning benchmark",
+    "Exploring high PT experimental charges through the lens of phase maps geology",
+    "SF Bay Area Coda Calibration Example Dataset seismology",
+]
+
+
+class EyeImagingClassifier:
+    """Eye imaging dataset classifier using SetFit few-shot learning.
+
+    Classifies metadata records into 4 classes:
+      - EYE_IMAGING: Actual eye imaging datasets (fundus, OCT, OCTA, etc.)
+      - EYE_SOFTWARE: Code, tools, models for eye imaging (no actual data)
+      - EDGE_CASE: Eye research papers, reviews, borderline items
+      - NEGATIVE: Unrelated domains
+
+    Usage:
+        >>> from envision_classifier import EyeImagingClassifier
+        >>> clf = EyeImagingClassifier()
+        >>> clf.classify("Retinal OCT dataset for diabetic retinopathy")
+        {'label': 'EYE_IMAGING', 'confidence': 0.999, 'probabilities': {...}}
+    """
+
+    LABELS = LABELS
+
+    def __init__(self, model_path=None, device=None):
+        """Load a trained classifier model.
+
+        Args:
+            model_path: Path to saved model directory, HuggingFace repo ID,
+                       or None to auto-download from HuggingFace.
+            device: "cuda", "cpu", or None for auto-selection.
+        """
+        if device is None:
+            device = self._select_device()
+        self._device = device
+
+        if model_path is None:
+            model_path = self._download_model()
+        else:
+            model_path = Path(model_path)
+
+        self._load_local(model_path)
+        self._base_model_name = None  # set during train()
+
+    def _download_model(self):
+        """Download model from HuggingFace Hub, using cache."""
+        from huggingface_hub import snapshot_download
+
+        local_dir = snapshot_download(
+            repo_id=HF_MODEL_REPO,
+            cache_dir=None,  # uses default HF cache
+        )
+        return Path(local_dir)
+
+    def _load_local(self, model_path):
+        """Load model from local directory."""
+        from sentence_transformers import SentenceTransformer
+        import joblib
+
+        model_path = Path(model_path)
+        self._encoder = SentenceTransformer(str(model_path), trust_remote_code=True)
+        self._encoder = self._encoder.to(self._device)
+        self._head = joblib.load(model_path / "model_head.pkl")
+
+    @staticmethod
+    def _select_device():
+        """Auto-select best available GPU or fall back to CPU."""
+        import subprocess
+
+        try:
+            result = subprocess.run(
+                [
+                    "nvidia-smi",
+                    "--query-gpu=index,memory.free",
+                    "--format=csv,noheader,nounits",
+                ],
+                capture_output=True,
+                text=True,
+            )
+            gpus = [
+                (int(line.split(",")[0]), int(line.split(",")[1]))
+                for line in result.stdout.strip().split("\n")
+            ]
+            best_gpu = max(gpus, key=lambda x: x[1])[0]
+            os.environ["CUDA_VISIBLE_DEVICES"] = str(best_gpu)
+        except Exception:
+            os.environ.setdefault("CUDA_VISIBLE_DEVICES", "0")
+        return "cuda" if torch.cuda.is_available() else "cpu"
+
+    def classify(self, metadata):
+        """Classify a single metadata record.
+
+        Args:
+            metadata: dict with 'title', 'description', and/or 'keywords',
+                     or a plain text string.
+
+        Returns:
+            dict with 'label', 'confidence', and 'probabilities'.
+        """
+        if isinstance(metadata, str):
+            text = metadata
+        else:
+            text = self.extract_text(metadata)
+        return self._predict_batch([text])[0]
+
+    def classify_batch(self, records, batch_size=16):
+        """Classify multiple metadata records.
+
+        Args:
+            records: list of dicts (with title/description/keywords) or strings.
+            batch_size: batch size for inference.
+
+        Returns:
+            list of classification result dicts.
+        """
+        texts = []
+        for r in records:
+            if isinstance(r, str):
+                texts.append(r)
+            else:
+                texts.append(self.extract_text(r))
+
+        all_results = []
+        for i in range(0, len(texts), batch_size):
+            batch = texts[i : i + batch_size]
+            all_results.extend(self._predict_batch(batch))
+        return all_results
+
+    def _predict_batch(self, texts):
+        """Run prediction on a batch of text strings."""
+        import numpy as np
+
+        embeddings = self._encoder.encode(texts, convert_to_numpy=True)
+        predictions = self._head.predict(embeddings)
+        probabilities = self._head.predict_proba(embeddings)
+
+        results = []
+        for i in range(len(texts)):
+            pred = predictions[i]
+            probs = probabilities[i]
+
+            if isinstance(pred, (int, float, np.integer)):
+                pred_int = int(pred)
+            else:
+                pred_int = {
+                    "NEGATIVE": 0,
+                    "EDGE_CASE": 1,
+                    "EYE_SOFTWARE": 2,
+                    "EYE_IMAGING": 3,
+                }.get(str(pred), 0)
+
+            label = self.LABELS[pred_int]
+
+            results.append(
+                {
+                    "label": label,
+                    "confidence": float(max(probs)),
+                    "probabilities": {
+                        "NEGATIVE": float(probs[0]),
+                        "EDGE_CASE": float(probs[1]),
+                        "EYE_SOFTWARE": float(probs[2]),
+                        "EYE_IMAGING": float(probs[3]),
+                    },
+                }
+            )
+        return results
+
+    @classmethod
+    def train(cls, output_dir=None, device=None, base_model_name=None,
+              num_epochs=2, batch_size=16):
+        """Train a new classifier from the built-in training data.
+
+        Args:
+            output_dir: Directory to save the model.
+            device: Device for training. None for auto-selection.
+            base_model_name: Sentence transformer backbone model name.
+                Defaults to BASE_MODEL_NAME (sentence-transformers/all-mpnet-base-v2).
+            num_epochs: Number of training epochs (default: 2).
+            batch_size: Training batch size (default: 16).
+
+        Returns:
+            EyeImagingClassifier instance with the trained model.
+        """
+        from setfit import SetFitModel, Trainer, TrainingArguments
+        from datasets import Dataset
+
+        if device is None:
+            device = cls._select_device()
+
+        if base_model_name is None:
+            base_model_name = BASE_MODEL_NAME
+
+        if output_dir is None:
+            output_dir = Path.cwd() / "models" / "setfit_v7"
+        else:
+            output_dir = Path(output_dir)
+
+        train_texts = (
+            EYE_IMAGING_EXAMPLES
+            + EYE_SOFTWARE_EXAMPLES
+            + EDGE_CASE_EXAMPLES
+            + NEGATIVE_EXAMPLES
+        )
+        train_labels = (
+            [3] * len(EYE_IMAGING_EXAMPLES)
+            + [2] * len(EYE_SOFTWARE_EXAMPLES)
+            + [1] * len(EDGE_CASE_EXAMPLES)
+            + [0] * len(NEGATIVE_EXAMPLES)
+        )
+
+        train_dataset = Dataset.from_dict(
+            {
+                "text": train_texts,
+                "label": train_labels,
+            }
+        )
+
+        print(f"Training SetFit model with {base_model_name}")
+        print(f"  Training examples: {len(train_dataset)}")
+        print(f"  Device: {device}")
+
+        model = SetFitModel.from_pretrained(
+            base_model_name,
+            labels=LABELS,
+            device=device,
+            trust_remote_code=True,
+        )
+
+        args = TrainingArguments(
+            output_dir=str(output_dir / "checkpoints"),
+            batch_size=batch_size,
+            num_epochs=num_epochs,
+            evaluation_strategy="no",
+            save_strategy="no",
+            logging_steps=50,
+        )
+
+        trainer = Trainer(
+            model=model,
+            args=args,
+            train_dataset=train_dataset,
+        )
+
+        print("Starting training...")
+        trainer.train()
+
+        output_dir.mkdir(exist_ok=True, parents=True)
+        model.save_pretrained(str(output_dir))
+        print(f"Model saved to: {output_dir}")
+
+        instance = cls(model_path=output_dir, device=device)
+        instance._base_model_name = base_model_name
+        return instance
+
+    @staticmethod
+    def extract_text(metadata):
+        """Compose classification text from metadata fields.
+
+        Args:
+            metadata: dict with 'title', 'description', and/or 'keywords'.
+
+        Returns:
+            Combined text string.
+        """
+        title = metadata.get("title", "")
+        desc = EyeImagingClassifier.strip_html(metadata.get("description", ""))
+        keywords = metadata.get("keywords", [])
+        if isinstance(keywords, list):
+            keywords = " ".join(keywords)
+        return f"{title} {desc} {keywords}".strip()
+
+    @staticmethod
+    def strip_html(text):
+        """Remove HTML tags from text."""
+        if not text:
+            return ""
+        clean = re.sub("<[^<]+?>", " ", text)
+        return unescape(clean).strip()

envision_classifier-0.1.0/envision_classifier/cli.py

@@ -0,0 +1,90 @@
+"""
+envision-classifier CLI
+
+Command-line interface for classifying eye imaging datasets.
+"""
+
+import json
+import sys
+
+import click
+
+
+@click.group()
+@click.version_option(package_name="envision-classifier")
+def cli():
+    """ENVISION: Eye imaging dataset classifier."""
+
+
+@cli.command()
+@click.argument("input_file", required=False, type=click.Path(exists=True))
+@click.option("--text", "-t", help="Classify a text string directly.")
+@click.option("--model", "-m", help="Path to trained model directory.")
+@click.option("--device", "-d", help="Device (cuda/cpu).")
+@click.option("--pretty/--compact", default=True, help="Pretty-print JSON output.")
+def classify(input_file, text, model, device, pretty):
+    """Classify metadata as eye imaging datasets.
+
+    Accepts a JSON file, --text string, or stdin.
+    """
+    from .classifier import EyeImagingClassifier
+
+    classifier = EyeImagingClassifier(model_path=model, device=device)
+    indent = 2 if pretty else None
+
+    if text:
+        result = classifier.classify(text)
+        click.echo(json.dumps(result, indent=indent))
+    elif input_file:
+        with open(input_file) as f:
+            data = json.load(f)
+        if isinstance(data, list):
+            results = classifier.classify_batch(data)
+        else:
+            results = classifier.classify(data)
+        click.echo(json.dumps(results, indent=indent))
+    elif not sys.stdin.isatty():
+        data = json.load(sys.stdin)
+        if isinstance(data, list):
+            results = classifier.classify_batch(data)
+        else:
+            results = classifier.classify(data)
+        click.echo(json.dumps(results, indent=indent))
+    else:
+        click.echo("Provide a JSON file, --text, or pipe JSON via stdin.", err=True)
+        raise SystemExit(1)
+
+
+@cli.command()
+@click.option("--output", "-o", help="Output directory for trained model.")
+@click.option("--device", "-d", help="Device (cuda/cpu).")
+def train(output, device):
+    """Train a new classifier from built-in training data."""
+    from .classifier import EyeImagingClassifier
+
+    classifier = EyeImagingClassifier.train(output_dir=output, device=device)
+    click.echo(f"\nModel ready. Labels: {classifier.LABELS}")
+
+
+@cli.command()
+def info():
+    """Display classifier information."""
+    from . import __version__
+    from .classifier import (
+        BASE_MODEL_NAME,
+        HF_MODEL_REPO,
+        LABELS,
+        EYE_IMAGING_EXAMPLES,
+        EYE_SOFTWARE_EXAMPLES,
+        EDGE_CASE_EXAMPLES,
+        NEGATIVE_EXAMPLES,
+    )
+
+    click.echo(f"envision-classifier v{__version__}")
+    click.echo(f"Base model:       {BASE_MODEL_NAME}")
+    click.echo(f"HuggingFace repo: {HF_MODEL_REPO}")
+    click.echo(f"Labels:           {', '.join(LABELS)}")
+    click.echo(f"Training data:    {len(EYE_IMAGING_EXAMPLES)} eye_imaging, "
+               f"{len(EYE_SOFTWARE_EXAMPLES)} eye_software, "
+               f"{len(EDGE_CASE_EXAMPLES)} edge_case, "
+               f"{len(NEGATIVE_EXAMPLES)} negative")

envision_classifier-0.1.0/pyproject.toml

@@ -0,0 +1,140 @@
+[tool.poetry]
+
+name = "envision-classifier"
+version = "0.1.0"
+description = "Few-shot classifier for detecting eye imaging datasets"
+
+packages = [{ include = "envision_classifier" }]
+
+license = "MIT"
+authors = [
+    "FAIR Data Innovations Hub <contact@fairdataihub.org>",
+    "James O'Neill <joneill@calmi2.org>",
+]
+
+readme = "README.md"
+homepage = "https://github.com/EyeACT/envision-classifier"
+documentation = "https://envision-classifier.readthedocs.io"
+repository = "https://github.com/EyeACT/envision-classifier"
+
+keywords = [
+    "eye imaging",
+    "ophthalmology",
+    "OCT",
+    "fundus",
+    "retina",
+    "machine learning",
+    "dataset discovery",
+    "classification",
+    "setfit",
+    "few-shot",
+    "fair-data",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Science/Research",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Natural Language :: English",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Medical Science Apps.",
+    "Topic :: Scientific/Engineering :: Image Processing",
+]
+
+[tool.poetry.dependencies]
+
+python = "^3.10"
+
+click = "^8.0"
+
+# ML/AI dependencies
+torch = ">=2.0.0"
+transformers = ">=4.35.0"
+setfit = ">=1.0.0"
+datasets = ">=2.14.0"
+scikit-learn = ">=1.3.0"
+huggingface-hub = ">=0.20.0"
+
+[tool.poetry.group.dev.dependencies]
+
+# Formatters
+black = "^22.1"
+tomli = "*"
+isort = "^5.10"
+
+# Linters
+mypy = "^1.0"
+pydocstyle = "^6.1"
+pylint = "~2.15"
+wrapt = "*"
+
+# Testing
+pytest = "^7.1"
+pytest-describe = "^2.0"
+pytest-cov = "^3.0"
+
+# Documentation
+mkdocs = "~1.3"
+pygments = "^2.11.1"
+
+# Tooling
+poethepoet = "^0.20.0"
+
+[tool.poetry.scripts]
+
+envision-classifier = "envision_classifier.cli:cli"
+
+[tool.poe.tasks]
+
+docs = "mkdocs serve"
+format_with_isort = "isort envision_classifier tests"
+format_with_black = "black envision_classifier tests"
+format = ["format_with_isort", "format_with_black"]
+typecheck = "mypy envision_classifier tests"
+pylint = "pylint envision_classifier tests --rcfile=.pylint.ini"
+precommit = ["format", "typecheck", "pylint"]
+test = "pytest -rx -W ignore::DeprecationWarning"
+
+[tool.black]
+
+quiet = true
+line-length = 100
+
+[tool.isort]
+
+profile = "black"
+line_length = 100
+
+[tool.mypy]
+
+ignore_missing_imports = true
+no_implicit_optional = true
+check_untyped_defs = true
+cache_dir = ".cache/mypy/"
+
+[tool.pytest.ini_options]
+
+addopts = """
+--strict-markers
+
+-r sxX
+--show-capture=log
+
+--cov-report=html
+--cov-report=term-missing:skip-covered
+--no-cov-on-fail
+"""
+
+cache_dir = ".cache/pytest/"
+
+markers = []
+
+[build-system]
+
+requires = ["poetry-core>=1.0.0"]
+build-backend = "poetry.core.masonry.api"