How hsg-aiml/remotesensingco2estimation works
Project type and purpose
• This is a research‐oriented machine‐learning codebase for estimating power‐plant CO₂ emissions from satellite imagery using a multitask deep network.
• It provides data loading, custom augmentations, model definition, training, and evaluation scripts.
File structure
• Top level:
– README.md (project overview)
– custom_augmentations.py (data‐augmentation routines)
– dataset_multitask.py (dataset class and loader)
– train_multitask.py (training entry point)
– eval_multitask.py (evaluation/inference entry point)
– models/ (model definition)
· model_multitask.py (multitask network architecture)
• No deployment or microservice code; this is a standalone training/evaluation pipeline.
README for additional insights
• Describes a three‐headed output: plume‐area mask, fuel type, power generation rate → CO₂ estimate.
• References external dataset on Zenodo.
• Emphasizes end-to-end training and physics-guided post‐processing.