We have a scene-graph generation task in Home Action Genome competition.
Sample code of dataloader
You can use sample scripts for preparing for competition.
Evaluation server open: end-April, 2022
Evaluation server close: 10 June, 2022
Report submission deadline: 15 June, 2022
Report submission page of HOMAGE competition (Microsoft CMT)
Workshop: 19 June, 2022
Challenge: Scene-graph Generation
We use scene graphs to describe the relationship between a person and the object used during the execution of an action. In this track, the algorithms need to predict per-frame scene graphs, including how they change as the video progresses. For this track, participants are also allowed to leverage audio information. External datasets for pre-training are allowed, but it needs to be clearly documented. Since there are multiple relationships between each pair of human and object, there is no graph constraint (or single-relationship constraint).
For evaluation of scene graph prediction, we use the evaluation metric as Scene graph classification (SGCLS).
The task is to predict object categories and predicate labels between the person and each object.
Participants can use input information as video, other modalities and ground truth boxes.
Evaluation metrics is recall@k, we compute the fraction of times the ground truth relationship triplets are predicted in the top k most confident relationships predictions in each tested frame.
We will use k=10, 20.
Evaluation code and submission Format
Here is the evaluation code for scene-gprah generation task.
Please download train data from "train data"
Please download test data from "test data for CVPR2022"
Please see "the Codalab page of HOMAGE competition"