SFC2Net
This repository implements SFC2Net proposed in the work:
High-Throughput Rice Density Estimation from Transplantation to Tillering Stages Using Deep Networks
Liang Liu, Hao Lu, Yanan Li, and Zhiguo Cao
Plant Phenomics, 2020
Model Structure
Qualitative Results
Installation
The code has been tested on Python 3.7.3 and PyTorch 1.3.1. Please follow the
official instructions to configure your environment. See other required packages
in requirements.txt
.
The RPC Dataset
- Download the Rice Plant Counting (RPC) test dataset from: BaiduYun (597 MB) (code: cirv) or OneDrive (597 MB)
- Unzip the dataset and move it into the
./data
folder, the path structure should look like this:
$./data/rice_datasets-test
├──── images
├──── label_mat
├──── test.txt
Inference
Pre-trained Model on RPC dataset
- Download the model from: BaiduYun (26.2 MB) (code: nhxk) or OneDrive (26.2 MB)
- Move the model into the folder, and the path structure should be:
$./snapshots/rice/sfc2net
├──── model_best.pth.tar
Evaluation
python hltest.py
Benchmark Results
Counting Results on the RPC Dataset
Method | Venue, Year | Pretrained | MAE | MSE | rMAE | r2 |
---|---|---|---|---|---|---|
MCNN | CVPR 2016 | No | 92.11 | 121.52 | 15.33 | 0.89 |
TasselNetV2 | Plant Methods 2019 | No | 59.39 | 95.80 | 7.86 | 0.91 |
CSRNet | CVPR 2018 | VGG16 | 49.22 | 74.58 | 7.47 | 0.91 |
BCNet | TCSVT 2019 | VGG16 | 31.28 | 49.82 | 4.76 | 0.96 |
SFC2Net | This Paper | MixNet-L | 25.51 | 38.06 | 3.82 | 0.98 |
Comparison of Different Backbones
Backbone | MAE | MSE | rMAE | r2 | #Param. | ImageNet Top-1 Acc. |
---|---|---|---|---|---|---|
ResNet18 | 31.82 | 66.80 | 4.66 | 0.93 | 12.6M | 69.8 |
ResNet34 | 34.42 | 61.58 | 4.95 | 0.94 | 22.7M | 73.3 |
ResNet50 | 30.94 | 67.52 | 4.45 | 0.92 | 44.5M | 76.2 |
ResNet101 | 35.53 | 56.26 | 4.99 | 0.95 | 63.5M | 77.4 |
ResNet152 | 32.20 | 67.77 | 4.71 | 0.93 | 79.2M | 78.3 |
EfficientNet-B0 | 36.65 | 70.74 | 5.30 | 0.92 | 8.1M | 77.3 |
EfficientNet-B1 | 27.51 | 42.80 | 4.14 | 0.97 | 13.1M | 79.2 |
EfficientNet-B2 | 30.54 | 53.65 | 4.48 | 0.95 | 15.5M | 80.3 |
EfficientNet-B3 | 30.76 | 54.52 | 4.44 | 0.95 | 21.5M | 81.7 |
EfficientNet-B4 | 28.06 | 52.17 | 4.24 | 0.95 | 35.3M | 83.0 |
EfficientNet-B5 | 27.36 | 41.91 | 4.16 | 0.97 | 56.8M | 83.7 |
EfficientNet-B6 | 29.96 | 50.03 | 4.42 | 0.96 | 81.7M | 84.2 |
EfficientNet-B7 | 27.15 | 40.79 | 3.96 | 0.97 | 127.8M | 84.4 |
VGG16 | 30.67 | 57.53 | 4.51 | 0.95 | 15.7M | 71.6 |
MixNet-L | 25.51 | 38.06 | 3.82 | 0.98 | 8.3M | 78.9 |
Citation
If you find this work or code useful for your research, please cite:
@article{liu2020high,
title={High-Throughput Rice Density Estimation from Transplantation to Tillering Stages Using Deep Network},
author={Liu, Liang and Lu, Hao and Li, Yanan and Cao, Zhiguo},
journal={Plant Phenomics},
year={2020}
}
Permission
The code and data are only for non-commercial purposes. Copyrights reserved.
The training set of the RPC dataset is made available upon request. Contact: Hao Lu (poppinace@foxmail.com)