arunindiran/biomass repository overview

⁠Tree Biomass Predictor

Aims to predict tree biomass from the image using a hybrid deep learning model. The object detection model bounds the tree trunk and the depth extraction model estimates the relative disparity map. The combination of results from both models assists to crop, map and calculate relative distance from the depth map of tree image. The polynomial regression model to convert the relative distance to the absolute distance of the tree from the mobile camera. The metadata such as focal length(F), image width(W) of image and absolute distance (D) is applied on the below formulae to estimate GBH of tree

The application was wrapped into the streamlit app for demo. The application request height and tree image (with exif data) as input for biomass prediction

⁠Usage

1. Pull the repository from dockerhub

docker pull arunindiran/biomass:latest

1. Run the docker image as a container with port forwarded to 8501

docker run -p 8501:8501 arunindiran/biomass

1. Open the browser with URL http://localhost:8501⁠

⁠To ensure

• Proper internet connection to hit dockerhub and torch hub repository
• The initial run will consume more time than subsequent run as the models are downloaded and cached on the initial run
• Make sure images with exif tags(focal length and width) are retained while uploading

⁠prerequisite

• docker
• python pip

⁠Example output