Satellite Image Analysis: Diffusion & YOLOv8 Training

Stable Diffusion and Yolov8 training of satellite images
-----------------------
Stable Diffusion
Milestone 1: Modify the stable diffusion pipeline to work with a specific repository.
Milestone 2: Optimise the pipeline to work with 200 .tif images and generate a batch of 200 and 400 images with a given text prompt.

Deliverables: Zip files of a 2 datasets with 200 and 400 images. Metrics folder for the model.

----------------
YOLOv8
--Mileston1--

For YOLOv8 I have the code that works but it is not trained. For this
1.Baseline training: Train YOLOv8 on only 160 real images (the images have .npy or .geojson labels for each image)
2. Generate pseudo-labels: Use this initial model to predict on the batch of synthetic data
3. Filter confident predictions: Keep only high-confidence pseudo-labels (e.g., confidence > 0.7)

--Milestone 2 -- if the confidence predictions are good then we move to this step.
4.Augmented training: Retrain model on:
Original 140 real images
Synthetic images with pseudo-labels
5. Final validation: Test ONLY on the 40 real images

Deliverables: Zip files of a metrics folder for the model, the data with the generated pseudo-labels