AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

DropLoss for Long-Tail Instance Segmentation

Ting-I Hsieh, Esther Robb, Hwann-Tzong Chen, Jia-Bin Huang

AAAI 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Amodal and open-world instance segmentation · 36 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace BCE with DropLossMask R-CNN, R-50-FPN · LVIS v0.5 valAP (%): 21.5 -> 25.5Improved
Replace EQL with DropLossMask R-CNN, R-50-FPN · LVIS v0.5 valAP (%): 23.8 -> 25.5Improved
Replace EQL with DropLossMask R-CNN, R-50-FPN · LVIS v0.5 valAP_r: 8.5 -> 13.2Improved
Replace EQL with DropLossMask R-CNN, R-50-FPN · LVIS v0.5 valAP_f: 28.3 -> 27.3Hurt
Replace EQL with DropLossMask R-CNN, R-101-FPN · LVIS v0.5 valAP (%): 26.2 -> 26.9Improved
Replace EQL with DropLossCascade R-CNN, R-50-FPN · LVIS v0.5 valAP (%): 24.2 -> 25Improved
Replace EQL with DropLossCascade R-CNN, R-101-FPN · LVIS v0.5 valAP (%): 25.4 -> 26.4Improved
Add Repeat Factor SamplingDropLoss with Mask R-CNN, ResNet-50 · LVIS v0.5 valAP (%): 25.5 -> 26.4Improved
Add Repeat Factor SamplingDropLoss with Mask R-CNN, ResNet-50 · LVIS v1.0 valAP (%): 19.8 -> 22.3Improved
Use BEQL with logarithm base b=4 instead of DropLoss (ablation)Mask R-CNN, ResNet-50 · LVIS v0.5 valAP (%): 25.5 -> 25.2Hurt
Use BEQL with logarithm base b=5 instead of DropLoss (ablation)Mask R-CNN, ResNet-50 · LVIS v0.5 valAP (%): 25.5 -> 25.1Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy