Submerged – flora aimbire

SUBMERGED

master's degree project

SUBMERGED

MASTER'S DEGREE

PROJECT

@2018-2021

An immersive, real-time interactive art installation powered by computer vision and neural networks (2018–2021).

Bergerson Jewels’ Euphoria campaign unveils their stunning, fan-inspired collection with a narrative of transformation and vibrant self-expression. This seasonal campaign featured the design of a meticulously crafted catalog and its showcase within their exclusive bi-annual magazine. A beautifully produced social media film, co-created with a shared creative vision, brought the collection to life, capturing the interplay of motion, brilliance, and elegance. Rooted in Bergerson’s dedication to luxury and innovation, Euphoria invites its audience to embrace the dazzling beauty of a new journey.

Role:

Creator / Researcher / Developer / Artist

Visual concept, system architecture, motion capture testing,

brushstroke algorithms, interface design, and visual refinement.

Thanks to:

Dr. Roberto Zanetti (PUC-PR | UTFPR)

CAPES (financing)

Programmer:

Ms. Vinicius Ramalho

Visual reference for concept of Submerged, by photographer Guilherme Bitencourt @bitencourttavar

the vision

Before A.I. became the buzzword of the decade, Submerged was conceived to dissolve the boundaries between human expression and machine perception. The goal? To let the human body become brush, gesture become paint, and code become canvas.

As a creative trained in visual arts and architecture, I was frustrated with the rigid walls between art and technology. This project emerged from that friction—a desire to build art that moved with the viewer. A canvas that existed only through interaction, and died without it.

the concept

Inspired by works like Text Rain (Camille Utterback) and teamLab’s immersive environments, Submerged invites a solitary viewer into a responsive space.

Using Keypoint R-CNN, the system tracks body movements and transforms them into fluid, expressive brushstrokesprojected in real-time across the room.

The result? A digital painting that exists only in the now—generated, shaped, and destroyed by the viewer’s own presence and motion.

the tech

At the time, access to hardware and data was limited. I developed the system using:

Convolutional Neural Networks (CNNs) for pose estimation
Computer Vision pipelines using Keypoint detection (pre-Mediapipe era)
Custom algorithms for dynamic brushstroke generation based on movement, speed, and vector magnitude

The system was heavy on GPU/CPU resources, yet functional. Its performance pushed the limits of my personal hardware at the time—often compromising the aesthetic and conceptual integrity.

Flora-Aimbiré_SUBMERGED–A-HUMAN-COMPUTER-INTERACTIVE-ART-INSTALLATION-USING-COMPUTER-VISION-THROUGH-CONVOLUTIONAL-NEURAL-NETWORK-BODY-TRACKING

The moment

art and math

colide.

Testing the system with professional dancers allowed precise observation of keypoint displacement. Here, the leg movement between two frames is analyzed to calculate vector direction and speed, which in turn generated unique brushstroke effects in real-time.

Architecture testing, Sequence of 14 frames with CNN keypoint capture, brushstroke and color effect aplication.

This image captures a working test of Keypoint R-CNN identifying full-body landmarks in real time. Each dot represents a tracked joint—allowing the system to calculate displacement vectors, detect motion patterns, and convert the dancer’s presence into brushstroke logic.

It’s the invisible choreography that makes Submerged come alive.

Dance

Into

Data

In this sequence, a ballerina’s movement is captured in real-time by the Submerged system.

As she twirls and leaps, Keypoint R-CNN tracks her motion, translating it into a cascade of generative brushstrokes projected behind her. Each gesture becomes pigment; each step, a fleeting painting.

This test validated the core interaction concept: art that breathes with the body.

why it still matters

In 2018, there were no generative tools, no Stable Diffusion, no Runway ML. What existed were open-source libraries, hand-crafted models, and late-night debugging sessions.
I wasn’t just using AI—I was trying to create an emotional, embodied experience through it.

Unfortunately, I had to halt development due to technical limitations, lack of financial support, and the 2020 pandemic. But in retrospect, Submerged wasn’t just a prototype. It was a prelude—a signal of where art and interaction were going.

possibilities for today

With today’s tools (ComfyUI, Runway, Unity AI, LoRA training), Submerged could evolve:

From a Kinect room to a wearable AR experience
From real-time projection to interactive installations using stable, real-time ML pipelines
From a single-user canvas to a shared social installation, combining multiplayer motion and neural rendering

It was never just about tech—it was about creating awe through art that breathes with the viewer. And that hasn’t changed.

Submerged was never just about using technology—it was about feeling through it.
It was about exploring how our bodies can paint, how machines can translate presence into poetry, and how art doesn’t have to be static to be sublime.

I built this project before the tools we now take for granted even existed. I was chasing a vision with intuition, research, and raw curiosity. And while I couldn’t take it as far as I wanted then, the concept remains deeply relevant—perhaps even more so now.

Because in a world overwhelmed by generative content, what stands out is not what machines can do alone, but what they can co-create with us.

Submerged is the foundation of my ongoing exploration into human-machine creative symbiosis. And this is only the beginning.

← Previous

→ Next