Supreme: Fine-grained Radio map Reconstruction via Spatial-Temporal Fusion Network

In this paper, we propose a fine-grained radio map reconstruction framework, called Supreme, based on crowd-sourced data in an image super-resolution manner.

Paper

Kehan Li, Jiming Chen, Baosheng Yu, Chao Li, Zhangchong Shen, Shibo He. "Supreme: Fine-grained Radio map Reconstruction via Spatial-Temporal Fusion Network.", Accept by IPSN 2020.

Requirements

Supreme uses the following dependencies:

• Pytorch 0.4.3 and its dependencies
• Numpy, Scipy and Pandas
• CUDA 10.0 or latest version. And cuDNN is highly recommended

Model Training

Main arguments:

• n_epochs：number of epochs of training
• batch_size: training batch size
• lr: learning rate
• n_residuals: number of residual blocks
• base_channels: number of feature maps
• img__width: radio map width
• img_height: radio map height
• depth: number of historical radio maps
• channels: number of radio map channels
• sample_interval: interval of validation
• zoom: upscale factor of radio maps
• ext_flag: whether to use external factor

Examples on model training:

• Training Supreme with default setting:

python train.py --ext_flag

• Training Supreme with given setting (with external factor):

python train.py --n_residuals=16 --base_channels=64 --depth=6 --ext_flag

• Training Supreme without external factor

python train.py --n_residuals=16 --base_channels=64 --depth=6

Model Test

To test trained model, following code can be used:

python test.py --n_residuals=16 --base_channels=64 --depth=6

Dataset

The dataset is divided into train, validation and test set with a ratio 4:1:1.