GETTING_STARTED.md

Getting Started (Graph Generation)

Task-specific FCs

To run the graph generation baseline using task-specific FCs, run the script:

python run_easg.py annts_in_new_format data your_output_folder The data folder should contain

• Verb features
• RoI object features for train set
• RoI object features for validation set

This script also relies on the annotations in the format adjusted for SGG purposes stored in easg-generation/annts_in_new_format/.

The code for extracting object and verb features is provided in easg-generation/utils/. Verb features are based on Ego4D SlowFast features. Object features are based on detections of Faster-RCNN model, trained on our Ego4D-EASG dataset

STTran

This baseline builds on the original STTran repository. Please follow the instructions for requirements or other preparation.

To run the graph generation baseline using STTran, run the script:

python train_with_EASG.py -mode $MODE -data_path $DATAPATH

MODE: edgecls, sgcls, or easgcls