One-Shot Quality Claims Did Not Survive a Second Seed
Two quality claims came from one shot each. A second seed moves the metric by more than the gap between the step counts, so we retracted both and kept the test that can rank them.
Posts in this category: 4
Two quality claims came from one shot each. A second seed moves the metric by more than the gap between the step counts, so we retracted both and kept the test that can rank them.
Prompts that opened with lighting dropped the subject in four of six shots. The fix we wrote names the subject and place first, but no re-run with the new wording is recorded, so the fix is untested.
Our final model, a six-model ensemble with 8-view test-time augmentation, scores 0.4188 panoptic quality in 5-fold validation and 0.36 on the public board, rank 730 of 1,549. Validation sits about 0.06 above the board, and we cannot yet say why.
Kaggle · part 2Eight raw hidden vectors did not improve accuracy in a same-model 120-task rerun.