Symmetrical Visual Contrastive Optimization: Aligning Vision-Language Models with Minimal Contrastive Images
A two-color argument: cyan means the matching image, orange the mismatched one, and that code runs unbroken from the prose highlights into every figure across alternating grey and white bands.
Checked against the rendered page.
Why it is here
Banded sections are matched to a two-color (blue/orange) semantic palette that runs from the text highlights into the figures, so the visual language stays consistent. Teaser and method figures are well integrated with the prose. The qualitative grid gets dense at the bottom.
Overview
Color Palette
Typography
Google Sans
Symmetrical Visual Contrastive Optimization: Aligning Vision-Language
Abstract · Method · Results · BibTeX — 0123456789
Noto Sans
Symmetrical Visual Contrastive Optimization: Aligning Vision-Language
Abstract · Method · Results · BibTeX — 0123456789
Spacing & Shape
Page Modules
Guidelines
More like this
GAMEBoT: Transparent Assessment of LLM Reasoning in Games
A tidy game-arena scoreboard: a mascot crest over a navy title, then cool grey-blue gradient bands holding news cards and violet-headed score tables.
InstructPart: Task-Oriented Part Segmentation with Instruction Reasoning
A lab notebook in grey and white stripes, where the title marks its own key words in teal and a dashed-panel kettle explains the task before you read a line.
Slamming: Training a Speech Language Model on One GPU in a Day
A one-chart argument: a big annotated scaling plot sits where the teaser video would be, then a grey abstract and a quiet grid of audio players tagged by GPU budget.
Peeking Ahead of the Field Study: Exploring VLM Personas as Support Tools for Embodied Studies in HCI
A field notebook taped open on a white desk: a tall soft-gray headline, a long line of sky-blue names, and a salmon-to-sand storyboard that explains the study in three panels.
Reflection Separation from a Single Image via Joint Latent Diffusion
A pane of night glass you can wipe clean with one drag.
SALAD: Skeleton-aware Latent Diffusion for Text-driven Motion Generation and Editing
A quiet typeset monograph, its grey and white plates turning one after another.