Modernizing Tree Detection with Advanced Object Detection Models (GSOC 2024)

[Om-Doiphode]

I am interested in this project. According to the project description, we are expected to update the existing 1 stage object detector used in DeepForest which is RetinaNet. RetinaNet uses Feature Pyramid Network (FPN) and focal loss to improve upon the class imbalance issue of 1 stage detector. Upon researching, I came across another model which is EfficientDet which is an extension of the EfficientNet family of convolutional neural networks. It is also a 1 stage detector and uses bidirectional FPN. My approach was to test this model for tree detection. Am I in the right direction? Also I have some questions like the project description states that we have to explore transformers for image representation, so won't it be computationally intensive to run transformers? They can also be slower during inference. Can you please elaborate more on these?