Skip to content

Latest commit

 

History

History
38 lines (23 loc) · 1.03 KB

README.md

File metadata and controls

38 lines (23 loc) · 1.03 KB
Raw

Bottom-up Attention Detector

The object detector is pre-trained on Visual Genome and can extract fine-grained visual (bbox) features and correlated object tags.

Installation

git clone https://github.com/airsplay/py-bottom-up-attention.git
cd py-bottom-up-attention

# Install python libraries
pip install -r requirements.txt
pip install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

# Install detectron2
python setup.py build develop

# or if you are on macOS
# MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ python setup.py build develop

# or, as an alternative to `setup.py`, do
# pip install [--editable] .

Feature Extraction Scripts for Open Images

  1. For Open Images: open images script
python demo/demo_oi_for_reddit.py

References

The code for retrieval model is implemented as a fork of "py-bottom-up-attention". More details please refer to their repo, many thanks for their work.