Table 1 Comparison of LACSS model and the baselines on the LIVECell dataset.

Cell lines	MaskRCNN (fully-supervised)				CenerMask (fully-supervised)				LACSS (semi-supervised)
	AP50	AP75	mAP	AJI	AP50	AP75	mAP	AJI	AP50	AP75	mAP	AJI
A172	0.74	0.36	0.38	0.59	0.72	0.30	0.35	0.59	0.68	0.29	0.33	0.54
BT474	0.76	0.44	0.43	0.56	0.78	0.43	0.43	0.52	0.68	0.36	0.37	0.53
BV2	0.85	0.60	0.53	0.68	0.85	0.45	0.46	0.67	0.88	0.58	0.52	0.63
Huh7	0.80	0.58	0.52	0.64	0.78	0.50	0.46	0.64	0.70	0.48	0.43	0.60
MCF7	0.74	0.36	0.38	0.53	0.76	0.33	0.37	0.57	0.71	0.27	0.33	0.53
SHSY5Y	0.61	0.16	0.25	0.49	0.61	0.14	0.24	0.50	0.43	0.06	0.14	0.40
SKOV3	0.87	0.60	0.54	0.73	0.86	0.53	0.49	0.71	0.82	0.48	0.46	0.66
SkBr3	0.91	0.80	0.65	0.81	0.90	0.80	0.66	0.79	0.93	0.81	0.65	0.75

All baseline models were trained with the full annotation, whereas LACSS models were trained using the incomplete annotation. Bold font highlights categories for which the LACSS outperformed the baselines.

Quick links

Search