Skip to content

Fine-tuned thin-plate spline motion model for manipulating social information in paper-wasp colonies

License

Notifications You must be signed in to change notification settings

PavlicLab/CVPR2024-CV4Animals2024-TPSM_for_Paper_Wasps

 
 

Repository files navigation

Output Videos

Our project showcases the remarkable capabilities of the TPS (Thin Plate Spline) algorithm in animating a wasp image based on driving videos. Below, you will find a series of output videos demonstrating the versatility and effectiveness of our models, with a special emphasis on the last GIF which represents the pinnacle of our custom training efforts.

Pre-Trained Model Outputs

  1. Wasp Animated by Human Motion
    In our first demonstration, we use a TPS pre-trained model to animate a wasp figure, identified by a distinctive blue dot, using a human driving video. This showcases the model's ability to transfer complex human movements to a significantly different creature like a wasp.

  2. Wasp Animated by Wasp Motion (Pre-Trained Model)
    Here, the same pre-trained TPS model animates our wasp figure using a driving video of another wasp. This test illustrates the model's initial testing on more congruent source and target pairs.

Custom Trained Model Output

  1. Enhanced Wasp Animation with Custom Training
    The culmination of our project thus far is presented in this output, where we've applied a TPS model that underwent custom training specifically tailored to the nuances of wasp motion. This GIF not only demonstrates a significant improvement in the fidelity and naturalism of the wasp animation but also highlights the effectiveness of our specialized training regimen. The precise movements, much closer to natural wasp actions, underscore the value of custom training for specific animation goals.

About

Fine-tuned thin-plate spline motion model for manipulating social information in paper-wasp colonies

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 81.0%
  • Python 19.0%