# peteanderson80/bottom-up-attention

Bottom-up attention model for image captioning and VQA, based on Faster R-CNN and Visual Genome

Repository: https://github.com/peteanderson80/bottom-up-attention
Canonical: https://ross.abutalabs.com/products/bottom-up-attention
Homepage: http://panderson.me/up-down-attention/
Language: Jupyter Notebook
License: MIT
License Family: permissive
Topics: vqa, visual-question-answering, captioning-images, faster-rcnn, caffe, image-captioning, mscoco, mscoco-dataset
Last push: 2023-02-03T04:27:32+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 3386, "days_push": 1307, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1469, forks 369 (observed 2026-08-28T04:04:49.057352+00:00)

## What it is
A bottom-up attention model based on Faster R-CNN with ResNet-101 trained on Visual Genome, producing features for salient image regions. These features serve as drop-in replacements for CNN features in attention-based image captioning and visual question answering models.

## Use cases
- generate bottom-up attention features for image captioning models
- extract salient region features for visual question answering
- train a Faster R-CNN model on Visual Genome object and attribute annotations
- download pretrained MSCOCO image features instead of building the model
- replicate state-of-the-art VQA challenge results from 2017
- improve captioning CIDEr and BLEU scores with object-level attention

## When to choose
- you need object-level region features for captioning or VQA research
- you want pretrained MSCOCO features without building Caffe code
- you are reproducing the Bottom-Up Top-Down attention paper

## When to avoid
- you need a modern maintained framework like PyTorch or Transformers
- you want an end-to-end captioning model (use the separate Up-Down-Captioner repo)
- your environment cannot support legacy Caffe-based training

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, computer-vision, image-processing, deep-learning
- domain: computer-vision, machine-learning, deep-learning
- platform: python
- tags: faster-rcnn, visual-genome, vqa, image-captioning, mscoco, caffe, attention-features, object-detection, natural-language-processing, linux, gpu

## Member repositories
- peteanderson80/bottom-up-attention (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:49.057352+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:34:54.299146+00:00, confidence not recorded.
  - readme: https://github.com/peteanderson80/bottom-up-attention (fetched 2026-08-28T04:04:49.057352+00:00, sha acf3d856e323)
  - homepage: http://panderson.me/up-down-attention/ (fetched 2026-08-29T11:42:56.475773+00:00, sha 7b422b4d8dce)
- Data as of 2026-08-30T08:39:29.467469+00:00.