Skip to main content

Table 7 Comparison with state-of-the-art counting approaches on the test set of WSC dataset. TasselNetv2 adopts an AlexNet-like architecture in Fig. 6 and is trained from scratch

From: TasselNetv2: in-field counting of wheat spikes with context-augmented local regression networks

MethodHenan Zhengzhou (2012–2013)Shandong Taian (2012–2013 Camera1)Overall#Parameters
MAERMSEMAERMSEMAERMSE
Segmentation method in [13]387.09436.84268.03345.78317.19386.22\(\times\)
CCNN [6]168.41214.4152.4072.78101.39149.91\(5.70\times 10^5\)
MCNN [5]149.44188.3458.8375.5097.08135.17\(1.33\times 10^5\)
CSRNet\(^\dagger\) [23]64.1988.9633.2646.1946.3267.63\(1.63\times 10^7\)
TasselNet [9]94.97137.2436.7957.3761.3599.27\(6.38\times 10^5\)
TasselNetv274.97113.2133.1249.2650.7980.66\(6.38\times 10^5\)
TasselNetv2\(^\dagger\)61.5787.6731.6247.5544.2767.47\(1.60\times 10^7\)
  1. \(^\dagger\) means the model is finetuned from the pretrained VGG16, and layer-by-layer settings can be found in Additional file. The best performance is italics