AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Exploring Vision-Language Models for Imbalanced Learning

Yidong Wang, Zhuohao Yu, Jindong Wang, Qiang Heng, Hao Chen, Wei Ye, Rui Xie, Xing Xie, Shikun Zhang

IJCV 2023 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 46 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 lightweight decoder with Softmax over linear probing (ablation)CLIP · ImageNet-LToverall accuracy (%): 67.38 -> 75.05Improved
add lightweight decoder with imbalanced algorithms (ablation)CLIP · iNaturalist18overall accuracy (%): 5.45 -> 69.82Improved
add decoder with Softmax instead of CoOp prompt tuningCLIP · ImageNet-LToverall accuracy (%): 60.30 -> 75.05Improved
add decoder with Softmax instead of CoOp prompt tuningCLIP · Places-LToverall accuracy (%): 26.10 -> 40.15Improved
add Balanced Softmax to decoder over plain Softmax (ablation)CLIP decoder · ImageNet-LToverall accuracy (%): 75.05 -> 79.03Improved
add imbalanced algorithm (Disalign) to decoder (ablation)CLIP decoder · Places-LToverall accuracy (%): 40.15 -> 48.43Improved
replace full finetuning with decoder plus imbalanced learningCLIP · iNaturalist18overall accuracy (%): 65.43 -> 73.24Improved
linear probing instead of zero-shotCLIP · ImageNet-LToverall accuracy (%): 70.54 -> 67.38Hurt
scale backbone from ViT-B16 (85M) to ViT-L14 (443M)CLIP decoder · all datasetsoverall accuracy (%)Improved
use Laion-CLIP (2B data) instead of CLIP (400M) as backboneCLIP decoder with imbalanced methods · iNaturalist18overall accuracy (%): 10%Hurt
add decoder to linear probingCLIP ViT · ImageNet-LTGPU memory: 13% to 17%Hurt
use CoOp instead of decoderCLIP · ImageNet-LT carboncarbon emitted: 67%Improved

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