Catellect-VL-2B
A Vision-Language Model for Edge-Based Feline Behavior Understanding
A compact vision-language model for structured behavior understanding and edge-oriented deployment.
- 40K
- Feline behavior samples
- 2B
- Parameters
- Vet
- Veterinarian-designed behavior schema
- Local
- Edge-oriented deployment
Evidence first.
Interpretation second.
We start with what can be observed—posture, ears, tail, face, interaction, and context—before inferring what it may mean.
- Stage 1
Structured behavior understanding
Observe and organize what is visibly present.
- Stage 2
Evidence-supported interpretation
Interpret only after the evidence has been assembled.
Research Fields
Our research spans individual-centric multimodal learning, edge intelligence, privacy and security, and user experience—from physical devices to digital interaction.
Individual-Centric
Multimodal Learning
Combining complementary signals with individual context for richer understanding over time.
- Temporal alignment of camera, audio, and IMU streams
- Respiration rate estimation from video under ambient motion
- Behavior sequence completion under partial occlusion
- Orientation-invariant activity recognition from a collar-worn IMU
View 1 more topicShow fewer topics
- Radar–vision fusion for sensing under low visibility
Edge Intelligence
Making intelligence work within the limits of everyday devices.
Privacy & Security
Privacy and protection as part of system design.
User Experience
From physical form and everyday use to digital interaction.
- Actuation and charging geometry for wider interaction coverage
- Battery runtime within an 8 g wearable mass budget
- Acoustic noise reduction in collar-mounted actuation