AI Paper Atlas

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

Confidence-aware Pseudo-label Learning for Weakly Supervised Visual Grounding

Yang Liu, Jiahua Zhang, Qingchao Chen, Yuxin Peng

ICCV 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Visual grounding transformers · 23 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
add Heuristic+ improvements over Pseudo-Q heuristic pipeline (ablation)CPL · RefCOCO+ valtop-1 accuracy (%): 38.88 -> 40.36Improved
add uni-modal real query propagation (ablation)CPL with Heuristic+ · RefCOCO+ valtop-1 accuracy (%): 40.36 -> 50.65Improved
add cross-modal verification module and selective grounding loss (ablation)CPL · RefCOCO+ valtop-1 accuracy (%): 51.08 -> 51.81Improved
add object-centric and relation-aware BLIP pipelines on top of H+ with real-query (ablation)CPL · RefCOCO+ val and testBtop-1 accuracy (%): val 50.65 -> 51.08; testB 45.52 -> 44.46Mixed
use BLIP alone to select proposals (ReBLIP) (ablation)visual grounding training · RefCOCO+ valtop-1 accuracy (%): 12.07 vs Pseudo-Q 38.88Hurt
increase number of object proposals from 4 to 10 per pipeline (ablation)CPL · RefCOCO+ valtop-1 accuracy (%): 40.83 -> 48.77Improved
increase pseudo-query number from 200 to all (ablation)CPL · RefCOCO+ valtop-1 accuracy (%): 49.94 -> 51.81Improved
swap in our detectors for Pseudo-Q (ablation)Pseudo-Q · RefCOCO+ valtop-1 accuracy (%): 38.88 -> 38.03No clear effect
fine-tune with 10% labeled dataCPL · RefCOCO+ valtop-1 accuracy (%): 50.65 -> 64.76Improved
freeze BERT during training (ablation)CPL (uni-modal) · RefCOCO+ valtop-1 accuracy (%): 50.65 -> 46.19Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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