README.md

Wheat Diseas Detection

from Scratch in /tensor_flow_wheat

-data found in sciebo in /Use cases/Landwirtschaft/wheat_disease_raw_data

Using Palma bash files

Train Keras model

1. edit train_keras.sh (uses train_keras_sequential.py)
2. sbatch path/to/bashfiles_palma/train_keras.sh

Test Keras model

1. edit test_keras.sh (uses test_keras_model.py)
2. sbatch path/to/bashfiles_palma/test_keras.sh

Convert Keras to tflite

1. edit convert_to_tflite.sh (uses convert_to_tflite.py)
2. sbatch path/to/bashfiles_palma/convert_to_tflite.sh

Local

cd ./tensor_flow_wheat

conda env create -f ./tf_2.13_env.yml -n ENV_NAME

conda activate ENV_NAME

python ./train_keras_sequential.py ( with set flags)

---

Edge Impulse Copy

data

Y_split already in /data. X_split can be downloaded from sciebo in Use cases/Landwirtschaft/edge_impulse_train_data

create conda env ( cd to edgeimpulse_copy)

On Palma

salloc --nodes 1 --cpus-per-task 8 -t 03:00:00 --mem 32G --gres gpu:1 --partition gpuexpress

module purge module load palma/2021a Miniconda3/4.9.2 CONDA_BASE=$(conda info --base) source $CONDA_BASE/etc/profile.d/conda.sh conda deactivate

conda env create -f ./wheat_dis_env.yaml -n ENV_NAME

conda activate ENV_NAME