Skip to content

deeplab-ai/zs-vrd-bmvc20

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 

From Saturation to Zero-Shot Visual Relationship Detection Using Local Context

Code for our BMVC 2020 paper.

Requirements

Tested with Python 3.5 and 3.6.

python3 -r requirements.txt

Setup

  1. Clone the repository
git clone https://github.com/deeplab-ai/zs-vrd-bmvc20.git
cd zs-vrd-bmvc20
  1. Setup data
python3 main_prerequisites.py
  1. Before training with synonyms, run notebook RelationshipSimilarities.ipynb!

Model zoo

The full list of tested models is included in common/models. There are two basic model categories:

  • (Scene Graph) Generators: models that use standard linear classifiers.
  • (Scene Graph) Projectors: models that construct linear classifiers using language.

Train/test a model

python3 main_research.py --model=MODEL --test_dataset=TEST_DATASET

See main_research.py for other input arguments.

Example:

python3 main.py --model=visual_spat_projector

will train a visual-spatial net with a local-context-aware classifier and test it on VRD.

Other useful flags:

  • compute_accuracy: compute accuracy instead of precision
  • use_weighted_ce: weight cross-entropy terms
  • use_merged: evaluate with merged synonym classes
  • test_dataset: specify a different dataset for testing (only valid for projectors)

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published