Table 1.

Precision, recall and F1-scores for scene graph generation on test split, transposed

Relation	Assist	Cement	Clean	Close	Cut	Drill	Hammer	Hold	Lying	Operate	Prepare	Saw	Suture	Touch	None	Average
Prec	0.64	0.86	0.49	0.97	0.36	0.97	0.86	0.78	1.00	0.85	0.72	0.86	1.00	0.61	0.98	0.80
Rec	0.88	0.93	0.78	0.91	0.72	0.97	0.95	0.90	0.98	0.79	0.90	0.82	0.58	0.74	0.99	0.86
F1	0.74	0.89	0.60	0.94	0.48	0.97	0.90	0.84	0.99	0.82	0.80	0.84	0.73	0.67	0.98	0.81

“Average” stands for macroaverage, which is the unweighted average over all classes. We use images, augmentations, linear loss weighting and PointNet++