AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

SiRi: A Simple Selective Retraining Mechanism for Transformer-Based Visual Grounding

Mengxue Qu, Yu Wu, Wu Liu, Qiqi Gong, Xiaodan Liang, Olga Russakovsky, Yao Zhao, Yunchao Wei

ECCV 2022 · Vision-language and multimodal › Multimodal LLMs · approach: Attention-based visual token pruning · 24 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 Selective Retraining (SiRi)MDETR* · RefCOCO+ testATop-1 accuracy (%): 75.65 -> 82.01Improved
use multi-task SiRi with an auxiliary decoder instead of single decoder (ablation)SiRi · RefCOCO+ testATop-1 accuracy (%): 82.01 -> 83.04Improved
add up to five selective retraining periods (ablation)MDETR* (multi-task) · RefCOCO+ valTop-1 accuracy (%): 72.29 -> 77.47Improved
add up to five selective retraining periods (ablation)MDETR* (multi-task) · RefCOCOg valTop-1 accuracy (%): 71.53 -> 77.39Improved
add SiRi when training on only 10% of dataMDETR* · RefCOCOg valTop-1 accuracy (%): 32.00 -> 61.58Improved
use learnable plus constant object queries in the two decoders instead of a single decoder (ablation)multi-task SiRi · RefCOCO+ valTop-1 accuracy (%): 70.93 -> 72.29Improved
use two identical query types in multi-task decoders (ablation)multi-task SiRi · RefCOCO+ valTop-1 accuracy (%)No clear effect
re-initialize visual backbone during retraining (ablation)SiRi · RefCOCO+ valTop-1 accuracy (%)Improved
add SiRiTransVG · RefCOCO+ valTop-1 accuracy (%): 66.39 -> 69.30Improved
add SiRiLAVT · RefCOCO+ valoIoU (%): 62.14 -> 62.86Improved
add SiRiLAVT · RefCOCO+ testBoIoU (%): 55.10 -> 55.03No clear effect
add SiRi retrainingMDETR phrase grounding · Flickr30kRecall@1 (%): 76.22 -> 78.63Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 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