jaywilson-hub / benchmark-on-so2satlcz42-dataset-a-simple-tour

LCZ classification from so2satLCZ42 dataset

basics and requirements

keras (tensorflow backend)

data: https://arxiv.org/abs/1912.12171

http://doi.org/10.14459/2018MP1454690

Folder Structure

pytorch-template/
├── train.py - main file for training (path to data needs to be set)
├── dataLoader.py - loading data from h5 files
├── model.py - architecture of sen2LCZ_drop
├── evaluation.py - evaluation of the trained models
├── lr.py - learning rate schedule
├── plotModel.py - plot the models
│
│
│
├── results/ - (temporary) results folder
│   ├── plotModel.py - plot the models
│   ├── modelS.py - select the models according to the setup(single task or mtl)
│   ├── model_sep_cbam.py - definition of the mtl framework
│   └── ...
│   
│    
├── img2map/ - predict using the trained models from s2 data
│   ├── img2lczMap_oneCity.py - read s2 data and predict and save the results in geotiff
│   ├── img2mapC4Lcz.py - functions for predictions
│   └── ...
│   
│
└── modelFig/ - figure of the model structure
    ├──  
    ├──
    └── ...       

Usage

img2map

• setting model path and image path, image data for test
• CUDA_VISIBLE_DEVICES=0 python img2lczMap_oneCity.py "../results/_32_weights.best.hdf5" "testData/00017_22007_Lagos"

train

• after setting the data path (in train.py): CUDA_VISIBLE_DEVICES=0 python train.py