Repository navigation
The new detection pipeline has been completed! Here are the trained YOLO models.
-
2015-trained-entire-year_differenced-norm-tiles_yolov8s-pt.pt
The original PyTorch model trained on the differenced images of 2015. This used all available data to train the model so this should be chosen for the best performance when rerunning detections. Download the converted ONNX model below when doing detections on Setonix. -
2015-trained-entire-year_differenced-norm-tiles_yolov8s-pt.onnx
The converted ONNX model that is optimised for CPU inference.
-
2015-trained-jul-to-dec_differenced-norm-tiles_yolov8s-pt.pt
The original PyTorch model trained on the differenced images of the second half of 2015. Use this to perform detections on the first half of 2015 for validation purposes since images from January to June are unseen by the model. Download the converted ONNX model below when doing detections on Setonix. -
2015-trained-jul-to-dec_differenced-norm-tiles_yolov8s-pt.onnx
The converted ONNX model optimised for CPU inference.
-
object_detection-2015_differenced_norm_tiles.zip
This is just the outputs from training the models above. -
2015-fireballs.txt
This is the list of fireballs from 2015 that were used to train the models above. Their before and after images were retrieved to create the differenced images.
all-standalone-fireballs-yolov8n-pt.pt
The PyTorch model trained on 6,554 fireballs but using the standalone RGB images. This is what was provided at the end of the Advanced Science project. Not really relevant now but added here just in case.