Question about relationship

[jungjun9150]

thank you for opening good code! I have a question. I implemented your code successfully and viusalized. but in the visualization image, there is no relationship name. so I modified your code to get triplet(object1, relationship, object2)

and I got the result. (I set the relation score threshold = 0.03 as compared to yours 0 sg_eval.py] line 45 (it may be some different) rel_score_threshold = float(0.03) sorted_inds = np.argsort(-scores) sorted_inds = sorted_inds[scores[sorted_inds] > rel_score_threshold] #[:100] )

, I used top_prediction object labels and got 923 top triplets I showed you the part of result in summary.

pant_wearing_jacket street_on_bus bus_on_street sign_on_street street_on_street pole_on_street street_on_sign pant_wearing_woman pant_on_street man_wearing_pant door_on_bus street_on_bag pole_on_sign woman_wearing_bag sign_on_bag bag_wearing_man jacket_on_street windshield_on_street jacket_wearing_woman jacket_wearing_man bus_on_bus sign_on_street window_on_bus door_on_street tire_on_bus pant_on_bus window_on_bus woman_on_street street_on_window man_on_street sign_on_bag pole_on_sign bus_on_window bus_on_sidewalk sidewalk_on_bus street_on_sidewalk tire_on_street bus_on_sign woman_wearing_woman man_wearing_woman bus_on_light window_on_street light_on_street window_on_door door_on_window street_on_window pant_wearing_person sign_on_sign tree_on_street sign_on_windshield light_on_bus bag_on_bus sign_on_light bus_on_door bag_wearing_jacket person_wearing_bag street_on_window bus_on_man roof_on_street person_wearing_jacket pole_on_bus bus_on_windshield bus_on_woman bus_on_jacket sign_on_window windshield_on_light window_on_sidewalk pole_on_bus sign_on_sidewalk sidewalk_on_pant sign_on_sign door_on_sidewalk street_on_building sidewalk_on_windshield bag_on_sidewalk sign_on_window man_on_bus sidewalk_on_sign window_on_door pant_on_bus pole_on_sidewalk street_on_person window_on_sign pant_on_roof window_on_bus sign_on_bus pant_wearing_pole hair_on_street tire_on_sidewalk sign_on_sign bus_on_bag sign_on_bus bus_on_woman man_wearing_person sidewalk_on_jacket woman_wearing_person sign_on_light pole_on_bag pole_on_window bus_on_windshield door_on_street bus_on_tree bag_on_window tire_on_door tire_on_bus street_on_bus sign_on_windshield hair_wearing_man sidewalk_on_sign bus_on_window hair_wearing_woman pant_on_building hair_wearing_pant pant_wearing_door pole_on_street window_on_sign pant_wearin

I think it is something different from your paper and the most relationship consist of 'on'. which variable I have to use to get relationship efficiently corresponding to top_predictions?

thank you Jianwei Yang.