Table 2.

Recall and precision of different models trained by k-d tree SVM-quantizer and SGD solver for different dictionary sizes (vocsize) and colour spaces (colour) based on a pooled dataset from Lab0d and Lab7d (temporal pooling).

ID	Voc size	Colour	Recall in %					Class mean in %
			BEMITA	ENCAFO	MACRPY	TRIAVA	BKGRND
163	200	Greyscale	80.00	92.51	93.85	77.41	99.16	88.59
164	200	RGB	84.84	92.37	97.09	76.72	99.16	90.04
165	200	HSV	81.89	92.51	97.41	72.45	99.16	88.69
169	500	Greyscale	77.05	92.51	95.79	78.79	99.16	88.66
170	500	RGB	81.68	92.37	97.09	76.58	99.16	89.38
171	500	HSV	80.63	92.23	96.76	69.01	99.16	87.56
			Precision in %
163	200	Greyscale	67.86	99.09	100.00	86.33	85.71	87.80
164	200	RGB	68.31	99.24	99.01	89.26	88.72	88.91
165	200	HSV	64.30	99.09	98.37	88.11	87.19	87.41
169	500	Greyscale	69.98	99.09	100.00	86.27	81.94	87.46
170	500	RGB	68.43	99.39	95.85	90.41	83.89	87.59
171	500	HSV	61.87	99.39	97.39	88.99	82.52	86.03