Skip to content

Latest commit

 

History

History
95 lines (75 loc) · 2.36 KB

OLD_README.md

File metadata and controls

95 lines (75 loc) · 2.36 KB
Raw

DeepDeconv

Deep learning based deconvolution algorithm implemented with a U-Net.

Download data

Download training data from:

https://figshare.com/articles/software/Deep_Deconvolution_for_Traffic_Analysis_with_Distributed_Acoustic_Sensing_Data/16653163

Virtualenv on Windows

Install virtualenv

pip install virtualenv

Create environment

virtualenv DeepDeconv

Activate environment

DeepDeconv\Scripts\activate

Install requirements

pip install -r requirements.txt

TEST MODEL

Testing the model:

python test.py --weights "/weights/best.ckpt"

TRAIN MODEL

Training the model:

python train.py --epochs 1000

In case "cannot be loaded because running scripts is disabled on this system":

Set-ExecutionPolicy Unrestricted -Scope Process

Fast Demo:

Download data form:

https://drive.google.com/drive/folders/1lKBVzk8I8hXu1jNqonX3EU2m1UF7nao3?usp=sharing

DAS chirp sin cambio de fase entre canales

python test.py --data data/CHIRP_DAS_NOFASE_data.h5 --weights /weights/200-epoch-chirp-single-channel/best.ckpt --kernel kernels/chirp_kernel.npy --act_function tanh

DAS chirp con aceleración entre canales

python test.py --data data/CHIRP_DAS_FASE_data.h5 --weights /weights/200-epoch-chirp-multi-channel/best.ckpt --kernel kernels/chirp_kernel.npy --act_function tanh

DAS de los autores

python test.py --weights /weights/1000-epoch-authors-integrado/best.ckpt --authors --kernel kernels/kernel.npy -ncc -pcc --act_function relu --integrate

DAS kernel flip sin integrar

python test.py --weights /weights/200-epoch-kernel-flip-sin-integrar/best.ckpt --authors --kernel kernels/kernel.npy -ncc --act_function relu

DAS kernel no flip sin integrar

python test.py --weights /weights/200-epoch-kernel-no-flip-sin-integrar/best.ckpt --authors --kernel kernels/kernel.npy -ncc -pcc --act_function relu

Figura 4, diferencia entre flip y no flip sin integrart en data de los autores

python utils/Figura_4.py -ncc --authors --act_function relu -pcc

Crear las figuras del entrenamiento a partir de los json

 python utils/ploting_training.py -if .\trainHistory\chirp-noflip.json -op C:\Users\Juan\Desktop\DeepDeconvV2\Figuras

Comparison chirp

python comparison_chirp.py --data data/CHIRP_DAS_NOFASE_data.h5 --weights /weights/200-epoch-chirp-single-channel/best.ckpt