Using only loss-scale balancing (log transformation) yields Δp = +0.06±0.09 on NYUv2.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Combining loss-scale and gradient-magnitude balancing achieves Δp = +1.15±0.16 on NYUv2.finding0.898
  Full DB-MTL ablation result.
• The logarithm transformation (loss-scale balancing) consistently outperforms IMTL-L on NYUv2, Cityscapes, Office-31, Office-Home.finding0.825
  Comparison of loss-scale balancing with IMTL-L.
• DB-MTL achieves ∆p = +1.15±0.16 on NYUv2, outperforming all baselines including state-of-the-artfinding0.823
  Primary empirical validation on scene understanding task
• DB-MTL with SegNet backbone achieves Δp = +8.91 on NYUv2, best among all methods.finding0.772
  Performance with a different backbone network.
• DB-MTL has similar per-epoch running time to gradient balancing methods on NYUv2, slower than loss balancing methods.finding0.763
  Computational efficiency comparison.
• On NYUv2, EW suffers a drop in surface normal prediction (mean angle error 23.57 vs STL 21.99, within 11.25° 35.04 vs 39.04).finding0.756
  Task balancing issue where surface normal prediction degrades under EW.
• DB-MTL achieves loss-scale balancing by performing logarithm transformation on each task loss, and rescales gradient magnitudes by normalizing all task gradients to comparable magnitudes using the maximum gradient norm.quote0.744
  Concise summary of the DB-MTL method from the abstract.
• At layer 0 α=5, detection-adjusted logit difference is +3.19 and control increase is +3.22, a difference of only 0.03 logitsfinding0.742
  Concrete numerical example showing detection and control are nearly identical at peak apparent accuracy