jaina1008/firemapping-model

Trained ResNet on Australian Bushfire linescan images to predict fire as a 0:1 and then predict the path of fire. This assists the firefighter department in allocating their limited manpower in areas where the fire will likely traverse.

Fire_Mapping_Project

Setting up the environment

1. Clone the repo.

git clone https://github.com/FoundingTitan/Fire_Mapping_Project.git

2. Copy the contents and paste it into the root of Google Drive.

/content/drive/MyDrive/Fire_Mapping_Project

3. Open a colab session at https://colab.research.google.com/

Make sure the session is set to GPU

Runtime > Change runtime Type > Hardware accelerator (from None to GPU)

4. Mount your google drive

from google.colab import drive
drive.mount('/content/drive')

4. In the first cell run

%cd /content/drive/MyDrive/Fire_Mapping_Project

Training

!python train.py

• Additional arguments can be supplied

Use of additional arguments

Example form of --arg option

!python -W ignore train.py --net attn_unet --epochs 200 > output_attn_unet.txt

• Specify model type --net

  • unet
  • attn_unet

• Specify number of epochs --epochs

  • Default 100
  • Any int value

• Batch-size --batch_size

  • Default 8
  • Any int value

• cutoff --cutoff

  • Default 0.3
  • Any float value

• learning rate --lr

  • Default 0.001
  • Any float value

• Print loss values every log_interval epochs --log_interval

  • Default 1
  • Any int value

• Transform data during training mode --transform_mode

  • Default Basic basic
  • Transform transform

• Transform type --transform_types

  • Default Crop crop
  • Horizontal Flip hflip
  • Vertical Flip vflip

• Set training seed --seed

  • Default 10
  • Any int value.

Alternative fast setup

• After putting the project in the root directory in the google drive + mounting.
• Go to Google colab and go File > Upload notebook > (change to the Upload tab) > Choose File.
• Choose the Main_train.ipynb or Main_Train_AUC.ipynb.
• Make sure the Session runtime is set to GPU (see above).
• Run all cells.

Adding augmented images

• Edit aug.py and add desired transforms at the top of the code. This utilizes the Albumentations package
• Determine file names during the saving stage located at the bottom of the code
• Run aug.py
• Copy and paste generated images located at augmented_images onto train_images
• Do the same for generated masks, from augmented_masks onto train_masks

Fire Prediction

• Open Wild_Fire.py with text editor to select image, image path and initial conditions
• Run Wild_Fire.py when satified.