AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Describing and Localizing Multiple Changes With Transformers

Yue Qiu, Shintaro Yamamoto, Kodai Nakashima, Ryota Suzuki, Kenji Iwata, Hirokatsu Kataoka, Yutaka Satoh

ICCV 2021 · Vision-language and multimodal › Human and embodied modeling · approach: Concept-guided image captioning · 127 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 the DUDA decoder with a transformer decoder (ablation)DUDA · CLEVR-Multi-ChangeBLEU-4: 76.1 -> 79.1Improved
replace the M-VAM decoder with a transformer decoder (ablation)M-VAM · CLEVR-Multi-ChangeBLEU-4: 62.9 -> 65.8Improved
concatenate before/after features over feature dimension rather than over patches (ablation)MCCFormers-D · CLEVR-Multi-ChangeBLEU-4: 80.1 -> 82.3Improved
concatenate before/after features over feature dimension rather than over patches (ablation)MCCFormers-S · CLEVR-Multi-ChangeBLEU-4: 80.6 -> 83.3Improved
use MCCFormers-S (single encoder) instead of D (dual) encoderMCCFormers · CLEVR-ChangeBLEU-4: 52.4 -> 57.4Improved
use MCCFormers-D instead of S encoderMCCFormers · Spot-the-Diff (multiple change)CIDEr: 18.2 -> 28.8Improved
use MCCFormers-D over DUDA on multi-change instancesMCCFormers-D · CLEVR-Multi-Change (4 changes)BLEU-4: 70.9 -> 80.6Improved
use MCCFormers-D over DUDA for sentence numberMCCFormers-D · CLEVR-Multi-Changesentence number accuracy: 83.3 -> 92.5Improved
use MCCFormers-D over DUDA on Spot-the-Diff multi-change BLEU-4MCCFormers-D · Spot-the-Diff (multiple change)BLEU-4: 5.4 -> 6.2No clear effect
use MCCFormers-D instead of M-VAM + RAF on Spot-the-Diff single changeMCCFormers-D · Spot-the-Diff (single change)BLEU-4: 11.1 -> 10.0No clear effect
use 2 encoder/decoder layers instead of 1 layer (1 head) (ablation)MCCFormers-D · CLEVR-Multi-ChangeBLEU-4: 59.0 -> 81.4Improved
use 4 layers instead of 2 layers (4 heads) (ablation)MCCFormers-D · CLEVR-Multi-ChangeBLEU-4: 82.3 -> 64.8Hurt

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