This is the repository for TasselNetv2, presented in:

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

Haipeng Xiong¹, Zhiguo Cao¹, Hao Lu¹, Simon Madec², Liang Liu¹, Chunhua Shen³

¹Huazhong University of Science and Technology, China

²INRA-EMMAH-CAPTE, 84914 Avignon, France

³The University of Adelaide, Australia

• matconvnet-1.0-beta22
• vlfeat-0.9.18

You can download the Wheat Spike Counting (WSC) dataset from: Google Drive

Pretrained models can be downloaded from: Google Drive

1. Download the code, data and model.

2. Organize them into one folder. The final path structure looks like this:

-->The whole project
    -->data
    -->model
        -->TasselNetv2_alex_patch64.mat
        -->TasselNetv2_vgg16_pre.mat
    -->vlfeat-0.9.18
    -->main.m
    -->paramInit.m
    -->genAnnotations.m
    -->hl_localreg.m
    -->hl_deploy_model.m
    -->get_stride.m

3. Run the following code to reproduce our results. Have fun:)

• To apply TasselNetv2, which is fast and accurate, please run: main(1). The result will be MAE: 50.16 and RMSE: 82.14
• To apply a VGG16 pretrained TasselNet, which is more accurate but much slower, please run: main(2). The reult will be MAE: 44.56 and RMSE: 68.32

You can refer to the offical link of MatConvNet for installation. After MatConvNet is installed, the opt.matconvnet_path variable in the paramInit.m file should be set to point to the corresponding path.