AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Simplified Concrete Dropout - Improving the Generation of Attribution Masks for Fine-grained Classification

Dimitri Korsch, Maha Shadaydeh, Joachim Denzler

IJCV 2025 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 2 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 original concrete dropout with the simplified sigmoid formulationFIDO · CUB-200-2011 SSR mask, ResNet50total variation: 39.74 -> 17.54Improved
replace original concrete dropout with the simplified formulationFIDO · CUB-200-2011 SDR mask, ResNet50total variation: 44.21 -> 22.72Improved
replace original concrete dropout with the simplified formulationFIDO · CUB-200-2011 Joint mask, InceptionV3 (iNaturalist2017)total variation: 28.66 -> 14.06Improved
use simplified concrete dropout with mini-batch size 8 instead of 32FIDO · CUB-200-2011 mask estimation, 100 stepscomputation time per image: 40 -> 11 secondsImproved
use simplified concrete dropout (smaller mini-batches)FIDO · Red-winged and Yellow-headed Blackbird masksIoU with ground-truth maskImproved
add FIDO joint-mask crop as test-time augmentationResNet50 · CUB-200-2011accuracy (%): 82.78 -> 84.17Improved
add simplified-CD FIDO crop instead of original FIDO cropResNet50 · CUB-200-2011accuracy (%): 84.17 -> 84.67Improved
add simplified-CD FIDO crop instead of original FIDO cropInceptionV3 (iNaturalist2017) · CUB-200-2011accuracy (%): 90.47 -> 90.51No clear effect
add random crop test-time augmentationInceptionV3 (iNaturalist2017) · CUB-200-2011accuracy (%): 90.32 -> 89.99Hurt
add center crop test-time augmentationResNet50 · CUB-200-2011accuracy (%): 82.78 -> 83.83Improved
add gradient-based crop test-time augmentationInceptionV3 · CUB-200-2011accuracy (%): 79.86 -> 80.76Improved
add BirdYolo detector crop test-time augmentationInceptionV3 · CUB-200-2011accuracy (%): 79.86 -> 81.12Improved

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