Table 5 Parameter settings of the CNN-based models for leaf counting.
From: A CNN-based model to count the leaves of rosette plants (LC-Net)
CNN model | Parameter setting |
|---|---|
VGG | The original input image has three channels and segmented mask has one channel. The size of both input image is (224, 224). For prediction, the Linear activation function is utilised. We have 45M trainable parameters for VGG model of Kumar et. al. and used loss function is half MSE and the optimizer is SGD with 0.0001 learning rate, 0.9 momentum (As mentioned in Kumar et. al.21) |
Alex Net | The original input image has three channels and segmented mask has one channel. The size of both input image is (224, 224). For prediction, the Linear activation function is utilised. We have 53M trainable parameters for Alex Net model of Kumar et. al. and used loss function is half MSE and the optimizer is SGD with 0.0001 learning rate, 0.9 momentum (As mentioned in Kumar et. al.21). |
Ubbans et. al. | The original input image has three channels and segmented mask has one channel. The size of both input image is (224, 224). For prediction, the Linear activation function is utilised. We have 45M trainable parameters for proposed model of Ubbans et. al. and the loss function is MSE (As mentioned in Ubbans et. al.31). |
Proposed LC-Net | The original input image has three channels and segmented mask has one channel. The size of both input image is (224, 224). For training, smooth L1 loss function has been used in Adam optimizer. We have 5M trainable parameters for our proposed LC-Net. |
