AI Paper Atlas

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

Improving Closed and Open-Vocabulary Attribute Prediction Using Transformers

Khoi Pham, Kushal Kafle, Zhe Lin, Zhihong Ding, Scott Cohen, Quan Tran, Abhinav Shrivastava

ECCV 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure · 14 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
pretrain on LSA then finetune on VAW (vs VAW only)TAP · VAWmAP: 65.4 -> 73.4Improved
use whole-image context with LSA pretraining (TAP vs RN50-Context)TAP · VAWmAP: 67.3 -> 73.4Improved
remove bounding boxes at test timeTAP · VAWmAP: 73.4 -> 68.9Hurt
use CLIP text embeddings instead of ViCo (ablation)OpenTAP · LSA common->rareAP overall: 7.98 -> 10.91Improved
use CLIP text embeddings instead of BERT (ablation)OpenTAP · LSA common->rareAP overall: 9.90 -> 10.91Improved
use CLIP text embeddings instead of Phrase BERT (ablation)OpenTAP · LSA common->rareAP overall: 10.26 -> 10.91Improved
add Flickr30K-Entities to VG training data (ablation)OpenTAP · LSA common->rareAP overall: 7.04 -> 8.16Improved
add MS-COCO (ungrounded) to training data (ablation)OpenTAP · LSA common->rareAP overall: 8.16 -> 9.56Improved
add Localized Narratives (mouse trace) to training data (ablation)OpenTAP · LSA common->rareAP overall: 9.56 -> 10.91Improved
combine object-aware with object-agnostic promptsCLIP · LSA commonAP overall: 2.64 -> 2.92Improved
replace CLIP with OpenTAP on unseen attributesOpenTAP vs CLIP combined · LSA common->rareAP unseen: 2.63 -> 5.37Improved
ensemble 1x, 1.25x, 1.5x crops of CLIP image embeddingsCLIP baselines · LSAmAP: +0.4 mAPImproved

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