About¶
License¶
imgad is distributed under the Apache 2.0 license. Models and datasets used with it keep their own licenses: check them before use, especially the research-only and non-commercial licenses common in medical imaging.
Acknowledgements¶
This work is partially supported by Agencia de Desarrollo Económico de La Rioja [ADER 2022-I-IDI-00015].
Releases¶
| Version | Highlights |
|---|---|
| 0.1.0 | Feature pipelines, classical detectors, Anomalib backend with anomaly maps, statistical benchmarking, markdown reports |
| 0.2.0 | Gated and private Hugging Face models; benchmark progress reporting |
| 0.3.0 | Parallel benchmarks with several models per GPU and memory checks |
| 0.4.0 | 16-bit, DICOM and NIfTI images with intensity rules; medical model zoo (imgad models) and timm extractor; documentation website |
| 0.5.0 | Ophthalmology: RETFound, RETFound-Green, VisionFM, RetiZero, FLAIR and UrFound in the zoo; raw checkpoints (Hub, URL, Google Drive) loaded into timm without executing code; ophthalmology guide |
| 0.6.0 | Zero- and few-shot detection with text prompts (zero_shot detector; FLAIR, RetiZero, BiomedCLIP, MedSigLIP, CLIP/SigLIP); few-shot protocol (k_shots); benchmarks continue when a run fails |
The full history is on GitHub releases.
Contributing¶
Issues and pull requests are welcome at github.com/joheras/imgad. The test suite runs offline:
Documentation lives in docs/ and is built with MkDocs Material: