Abstract
Collage is an art form that remixes visual elements to
create narratives.
However, it remains tedious and time-consuming to
select, cut out, and organize visual assets from an
unorganized set of images.
Informed by a formative study, we propose Collaposer, a
tool designed to automate the preparation of visual
assets for storytelling with collage.
Given a photo collection and a story description,
Collaposer selects and cuts out visual elements that are
both diverse in content and consistent with the story.
Specifically, objects are extracted via image tagging,
object detection, and segmentation, and then selected
and organized leveraging the reasoning capabilities of
LLM.
To facilitate asset selection, Collaposer provides a
curated overview that groups similar assets while
resizing them based on selection criteria.
A user study (N=12) demonstrated that Collaposer
produced rich story-relevant assets, facilitated
overview and navigation, and inspired users to create
diverse stories.
We further showcase our system's expressiveness in a
gallery.
Our pipeline consists of three stages.
The inputs include an image collection and a story
description.
In Stage I, valid visual elements are trimmed out and tagged
with an object name.
In Stage II, visual elements relevant to the story are
selected and clustered into semantic groups.
The elements classified as characters undergo part
segmentation and pose estimation for later manipulation.
In Stage III, the visual assets are visualized in a compact
view to facilitate navigation and composition.