From: High-throughput soybean seeds phenotyping with convolutional neural networks and transfer learning
Parameters | Values |
---|---|
Backbone layer | ResNet 50/101 |
Head layer | Faster R-CNN |
Input size | 256*256 / 512*512/1024*1024 |
Anchor ratio | [0.5, 1, 2] |
Learning rate | 0.001 |
Epoch | 40 |
RPN anchor scale | (8, 6, 32, 64-128) |
Pre-processing | Mean-subtraction |
Image resize mode | None |
Augmentation | LR, UD, Rotation, Brightness, Gaussian Blur |