Multi-Modal Foundation · MahmoodLab
CONCH v1.5
A pathology vision-language model for image feature extraction and zero-shot or few-shot tasks.
What CONCH v1.5 is
CONCH v1.5 is a pathology vision-language model developed by MahmoodLab for research in biology and medicine. It provides an updated checkpoint designed to interpret and extract features from histology imagery.
The model addresses computational pathology workflows where annotated training samples are limited or absent. By serving as a drop-in upgrade over earlier releases, it facilitates both zero-shot and few-shot pathology evaluations.
How it works
CONCH v1.5 incorporates a Vision Transformer (ViT) image encoder implemented in PyTorch using the timm library. The architecture features an updated, strengthened image-side encoder trained via self-supervised methods on an expanded multimodal pathology pretraining corpus.
Key facts
- Model type
- Pathology vision-language model
- Image architecture
- Vision Transformer (ViT)
- Primary pipeline
- Image feature extraction
- Frameworks
- PyTorch, timm
- Release date
- October 2024
What you can use it for
- Histopathology Feature Extraction
- Extract visual embeddings from histology image patches for downstream machine learning pipelines.
- Zero-Shot Pathology Evaluation
- Categorise and assess histological images directly using vision-language alignment without task-specific model fine-tuning.
- Few-Shot Pathology Classification
- Adapt the model to novel tissue types or disease categories using small sets of annotated pathology examples.
Run it here
We list this model but cannot run it yet: excluded (gated): Gated and CC-BY-NC-ND-4.0..
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Features a strengthened image-side encoder compared to the original CONCH release.
- Trained on an expanded pathology pretraining corpus.
- Functions as a drop-in upgrade for existing zero-shot and few-shot pathology analysis pipelines.
Watch out for
- Specific parameter counts and detailed benchmark evaluations are not specified in the model record.
- The exact composition, size, and source of the expanded pretraining corpus are not documented.
Inputs and outputs
You provide
Image
PNG, JPEG or TIFF. Sent to the model as-is.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · cc-by-nc-nd-4.0
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use CONCH v1.5 commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
What is CONCH v1.5?
It is an updated vision-language checkpoint developed by MahmoodLab for pathology and histology research.
How does CONCH v1.5 differ from prior checkpoints?
It incorporates an upgraded, strengthened image-side encoder and is trained on an expanded pretraining corpus.
What architecture does the vision component use?
The visual component uses a self-supervised Vision Transformer (ViT) supported by the timm library.
Written from CONCH v1.5's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.