Table 1 Parameters Setting of the existing CNN-based models used for leaf segmentation.
From: A CNN-based model to count the leaves of rosette plants (LC-Net)
CNN model | Parameter setting |
|---|---|
DeepLab V3+ | The input image has three channels and is (224, 224) in size. The backbone’s filters are identical to the original8. For prediction, the sigmoid activation function is utilised. DeepLab V3+ has 3.5M trainable parameters. |
SegNet | The input image has three channels and is (224, 224) in size. The backbone’s filters are identical to the original7. For prediction, the sigmoid activation function is utilised. We have 16.7M trainable parameters for the SegNet model. |
U-Net | The input image has three channels and is (224, 224) in size. The backbone and decoder filters are identical to the original9. Similar to the Fast FCN with PSP model, at the final level, the sigmoid activation function is applied in order to predict the class levels of each pixel. There are 31.0M trainable parameter present in the U-Net model used for the leaf segmentation. |
Refine Net | The input image has three channels and is (224, 224) in size. The backbone and decoder filters are identical to the original11. Similar to the above mentioned models, at the last layer, the final prediction is performed using the sigmoid activation function. The Refine Net model have 89.1M parameters that are needed to be trained. |
