Output Interpretation

Open ushahid opened this issue 5 years ago • 0 comments

Hi,

I have been able to successfully run the VG100k model on some random images and it produced the output in /Outputs/e2e_relcnn_X-101-64x4d-FPN_8_epochs_vg_y_loss_only/rel_detections.pkl however I am having some trouble interpreting the output. I understand that it gives me a list of dictionaries where each dictionary is associated with one image and has the following keys:

['sbj_boxes', 'sbj_labels', 'sbj_scores', 'obj_boxes', 'obj_labels', 'obj_scores', 'prd_scores', 'image']

Could you please explain what these mean and how can I go from here to visualizing the top K detected relationships? I tried using --vis command line option but it seems like that's a placeholder and hasn't been implemented yet.

Any help will be appreciated! If possible, can you provide the visualization code used in the paper?

Thanks, Usman

Mar 30 '20 01:03 ushahid