Home
Grounded Situation Recognition (GSR) - Interactive Visualization
Below we show results of our method and CoFormer on the SWiG testset, showcasing 100 randomly selected samples in a random order. For convenience, all visual results (grounded images and their predictions) are stored in a local folder, with a link provided for access. To explore the results navigate through the samples using the "Previous" and "Next" buttons. To view the input images along with results of both models, simply click the provided link to open the local file.
The predicted main activity and its corresponding semantic roles (arguments) are displayed beneath the image, with nouns labeled according to their specific roles within the activity. In our method, bounding boxes for the predicted AGENT are highlighted in pink, while bounding boxes for other predicted roles are shown in green and orange. For CoFormer, bounding boxes for the predicted AGENT are displayed in orange, and in the ground truth (GT), bounding boxes for the AGENT are highlighted in red.