AI Paper Atlas

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

Story Visualization by Online Text Augmentation with Context Memory

Daechul Ahn, Daneul Kim, Gwangmo Song, Seung Hwan Kim, Honglak Lee, Dongyeop Kang, Jonghyun Choi

ICCV 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure · 9 citations (OpenAlex)

Read the paper

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 partial-level memory connection (last layer only) (ablation)Single Directional Transformer · Pororo-SV (64x64)FID: 63.88 -> 59.05Improved
add attentively weighted memory (AWM) (ablation)PMA · Pororo-SV (64x64)FID: 59.05 -> 56.98Improved
add bi-directional training (ablation)PMA + AWM · Pororo-SV (64x64)Char. F1 / Frm. Acc.: FID 56.98 -> 54.69; Char. F1 50.15 -> 51.27; Frm. Acc. 22.01 -> 22.15Mixed
add online text augmentation (ablation)PMA + AWM + Bi. · Pororo-SV (64x64)FID: 54.69 -> 52.13Improved
use online instead of offline pseudo-text augmentation (ablation)CMOTA · Pororo-SV (64x64)FID: 54.51 -> 52.13Improved
use partial-level memory connection instead of all-level (ablation)CMOTA · Pororo-SV (64x64)FID: 61.23 -> 59.05Improved
increase resolution from 64x64 to 128x128CMOTA · Flintstones-SVFID: FID 36.71 -> 54.81; Char. F1 79.74 -> 86.44Mixed
use CMOTA instead of VP-CSV (character-centric module)CMOTA · Pororo-SV (64x64)Char. F1: 56.84 -> 53.25Hurt
use CMOTA (96.6M) instead of 1.3B StoryDALL-E fine-tuneCMOTA · Pororo-SVFID: 25.90 -> 55.26Hurt
add context memory (ablation)CMOTA · human study, temporal consistencypreference (%): 34.5% -> 61.2%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