golf-course-segmentation / evaluation.json
andrewluo's picture
Publish v4 segmentation model in Safetensors with inference documentation
43ee294 verified
Raw History Blame Contribute Delete
2.81 kB
{
"provenance": "Historical evaluations of the original v4 checkpoint; not rerun for this export.",
"checkpoint_epoch_zero_based": 12,
"checkpoint_metrics": {
"epoch": 13,
"train": {
"loss": 0.8420630561322281,
"green_iou": 0.7008157107509134,
"images": 4656,
"seconds": 279.1,
"class_iou": {
"green": 0.701,
"tee": 0.346,
"fairway": 0.648,
"bunker": 0.559,
"water": 0.029
}
},
"val": {
"loss": 0.930165826338124,
"green_iou": 0.6928858307247061,
"images": 573,
"seconds": 11.0,
"class_iou": {
"green": 0.693,
"tee": 0.331,
"fairway": 0.636,
"bunker": 0.563,
"water": 0.034
}
},
"lr": 5.171467231694576e-06,
"time": 1789530127.560203
},
"green_object_evaluations": {
"eval-negatives": {
"set": "negatives",
"threshold": 0.5,
"courses": 148,
"referenceGreens": 0,
"tiles": 148,
"candidatesPerKm2": 0.14,
"candidates": 2,
"recall50": 0.0,
"precision50": 0.0,
"recall25": 0.0,
"meanIoURefs": 0.0,
"meanIoUMatched": 0.0
},
"eval-osm_val": {
"set": "osm_val",
"threshold": 0.5,
"courses": 27,
"referenceGreens": 528,
"tiles": 823,
"candidatesPerKm2": 6.91,
"candidates": 537,
"recall50": 0.8788,
"precision50": 0.8641,
"recall25": 0.9015,
"meanIoURefs": 0.7303,
"meanIoUMatched": 0.7983
},
"eval-test_matched": {
"set": "test_matched",
"threshold": 0.5,
"courses": 99,
"referenceGreens": 1552,
"tiles": 2566,
"candidatesPerKm2": 7.14,
"candidates": 1729,
"recall50": 0.8827,
"precision50": 0.7924,
"recall25": 0.9072,
"meanIoURefs": 0.7283,
"meanIoUMatched": 0.8011
},
"eval-test_states": {
"set": "test_states",
"threshold": 0.5,
"courses": 40,
"referenceGreens": 805,
"tiles": 1374,
"candidatesPerKm2": 6.72,
"candidates": 871,
"recall50": 0.9031,
"precision50": 0.8347,
"recall25": 0.9205,
"meanIoURefs": 0.7616,
"meanIoUMatched": 0.823
}
},
"reviewed_green_evaluation": {
"set": "verified",
"threshold": 0.5,
"verified": 200,
"scored": 175,
"verdicts": {
"unclear": 20,
"osm": 17,
"model": 37,
"both": 119,
"reject": 5,
"neither": 2
},
"osmCorrectRate": 0.7771,
"packedModelCorrectRate": 0.8914,
"runMeanIoU": 0.8469,
"runRecall50": 0.9771,
"osmMeanIoU": 0.9335
},
"reviewed_reference_limitation": "136 of 175 scorable references reuse the OSM outline; this is not a fully independent human-traced benchmark."
}