Integrations

BNNR integrates with popular PyTorch vision stacks as an augmentation and analysis layer, not as a replacement training framework.

Overview

Integration	Status	Install	Entry point	Example
pytorch-grad-cam	Supported (core dependency)	pip install bnnr	ICD/AICD in your training loop	plugin_icd.md, gradcam_to_icd_loop.py
Ultralytics YOLO	Supported (adapter)	pip install "bnnr[ultralytics]"	UltralyticsDetectionAdapter	detection.md, ultralytics_yolo_quickstart.py
PyTorch Lightning	Reference	pip install pytorch-lightning	LightningAdapter	lightning_adapter.py
Hugging Face Accelerate	Reference	pip install accelerate	AccelerateAdapter	lightning_adapter.py

Runnable index: examples/integrations/README.md.

Positioning

pytorch-grad-cam answers where the model looks. BNNR adds ICD/AICD — augmentations that mask salient or background regions using those maps — plus optional BNNRTrainer branch search and bnnr analyze HTML reports. BNNR is not a second Grad-CAM library; it builds on grad-cam (see plugin_icd.md).

Ultralytics YOLO provides training and deployment via yolo train / YOLO.predict. BNNR wraps YOLO with UltralyticsDetectionAdapter for bbox-aware augmentation search, detection XAI, and structured reports — alongside your YOLO workflow, not instead of it.

Stable URLs for maintainers

Use these links in upstream docs or issues (replace main with a release tag when pinning versions):

Asset	URL
ICD plug-in guide	https://github.com/bnnr-team/bnnr/blob/main/docs/plugin_icd.md
Grad-CAM → ICD example	https://github.com/bnnr-team/bnnr/blob/main/examples/integrations/gradcam_to_icd_loop.py
Ultralytics quickstart	https://github.com/bnnr-team/bnnr/blob/main/examples/integrations/ultralytics_yolo_quickstart.py
This hub	https://github.com/bnnr-team/bnnr/blob/main/docs/integrations.md
Detection guide	https://github.com/bnnr-team/bnnr/blob/main/docs/detection.md
Citation (BibTeX)	https://github.com/bnnr-team/bnnr/blob/main/docs/citation.md

License note

BNNR is MIT. The Ultralytics repository is AGPL-3.0; using pip install ultralytics in your project is separate from forking Ultralytics. This documentation describes an optional adapter pattern only.

Citation

You use	Cite
BNNR (any feature)	BNNR software
ICD/AICD with gradcam saliency	ICD/AICD method paper, BNNR software, and pytorch-grad-cam
UltralyticsDetectionAdapter	BNNR software and Ultralytics per their license

BibTeX and plain-text formats: citation.md. GitHub also exposes this via CITATION.cff at the repository root.