A deep learning approach to track Arabidopsis seedlings’ circumnutation from time-lapse videos

Table 5 Effect of the search range on the tracking error and computation speed

Search range Video	\(101\times 101\)		\(201\times 201\)		\(401\times 401\)
Search range Video	Error (mm)	Speed (fps)	Error (mm)	Speed (fps)	Error (mm)	Speed (fps)
Val_1	1.82	93.6	2.54	73.9	0.32	31.5
Val_2	1.45	94.1	1.40	73.7	1.52	30.5
Test_1	2.53	91.3	1.68	73.3	0.91	30.5
Test_2	2.82	89.4	1.51	72.6	0.94	31.7
Test_3	1.37	92.2	1.85	74.7	1.43	31.7
Average	2.00	92.1	1.79	73.6	1.02	31.2

The speed is measured on a Nvidia RTX 2080 Max-Q laptop GPU. When running on an Intel i9-10980HK CPU, on average our program runs at 33fps, 16fps, and 6fps when the search range is set to \(101\times 101\), \(201\times 201\), and \(401\times 401\), respectively.
The bold values indicate that the corresponding algorithm (or model setting) performs the best in the comparison

ISSN: 1746-4811