What we're building

BNNR: PyTorch vision — train → explain → improve → prove. Core flows today: bnnr analyze (failure + XAI HTML), ICD/AICD augmentations, detection training via adapters.

Full product plan (by quarter): docs/roadmap.md

Next up (Q2–Q3 2026)

• Torchvision checkpoint → bnnr analyze (docs + example)
• Getting started links to the live sample HTML report
• GitHub templates for benchmark proposals and user showcases
• Multilabel CLI parity, Colab analyze notebook (Q3)
• Ecosystem: grad-cam tutorial, Ultralytics docs link when merged

Later

• bnnr compare (two checkpoints) — Q4
• Optional Fashion-MNIST / STL-10 benchmarks
• Detection analyze — 2027 H1, only after demand + stable classification path

Good first issues

Open tasks: https://github.com/bnnr-team/bnnr/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22

Epic (torchvision + analyze docs): #158 — child issues #159–#163

1. Comment "I'd like to work on this" on the issue
2. Wait for maintainer assignment
3. Fork → PR → CONTRIBUTING.md

We aim to reply and review within 48h on weekdays.

Other ways to participate

• Show and tell: share a run, report, or integration
• Questions: ask here before filing a bug
• Benchmarks: use the Benchmark issue template once chore: add GitHub issue template for benchmark contributions #161 is done

Try without training

Sample analyze HTML · pip install "bnnr[dashboard]" · python -m bnnr demo