In this project, we use drone images to find fledgling wildfires. The intent - find wildfires before they get 'wild'.
For this project, we obtained images from ieee-dataport . We did our own bounding box based data labelling using label studio and trained our model based on ~ 2000 images.
We ended up adapting resnet_v2 for our usecase. The result was a 0.82 mAP for a 0.5 IOU.
For a comprehensive overview of our building process, please refer to our Notion working Notes.
This projects was hosted by the Aggregate Intellect and done under the guidance of Andre Erler & Yan Nusinovich.