-
make repo self-contained: ship backbone.safetensors
-
Ship imagenet_classes.json so standalone ArgusEdge returns real ImageNet labels instead of class_0..999 placeholders.
-
README: drop training log reference
-
drop training-log process artifact
-
README: detection row 26.7 / 47.3 mAP
-
det_head: final 12-epoch checkpoint
-
README: detection row 26.0 / 46.5 mAP
-
det_head: swap to step036000
-
README: detection row 25.6 / 45.9 mAP
-
det_head: swap to step024000 (25.6 mAP)
-
README: update detection row to 24.1 / 43.4 mAP
-
det_head: swap to step010000 (24.1 mAP, stride-fix applied)
-
README: add detection row to results table
-
argus_edge.py: fix stride list to [8,16,32,64,128]; correct module docstring
-
Swap in improved det head (16.4 mAP vs 15.9)
-
detect: read num_classes from SplitTowerHead
-
from_pretrained: gracefully skip missing backbone.safetensors
-
Ship detection head
-
Ship detection head
-
Ship detection head
-
Ship depth + seg heads
-
Ship depth + seg heads
-
Ship depth + seg heads
-
Ship classifier head (val top-1 60.31 %)
-
Ship classifier head (val top-1 60.31 %)
-
Ship classifier head (val top-1 60.31 %)
-
Ship classifier head (val top-1 60.31 %)
-
Ship classifier head (val top-1 60.31 %)
-
Add model card
-
Remove premature README; repopulate after training
-
Initial README
-
initial commit
