Phosaiq: Photo mosaic art using quantum annealing

Symposium on Quantum–annealing Based HPC
February 15, 2023
@D–Wave Systems, Inc.

Transcript

1. Phosaiq: Photo mosaic art using quantum annealing Symposium on Quantum–annealing Based HPC February 15, 2023 @D–Wave Systems, Inc.
2. TEAM T-QARD-410 2 Tomomasa Hara Tohoku university Sigma-i Naoki Maruyama Tohoku university Sigma-i
3. What is Phosaiq? ▪ Phosaiq is a product which makes photo mosaic art by optimizing the arrangement of images ▪ Photo mosaic is a method of art in which many photos are assigned to divided area of the original image ▫ Partially just a series of images ▫ One picture as a whole 3
4. Feature of Phosaiq Hand-work ▪ Time-consuming and labor-intensive ▪ Handing over personal data 4 Simple algorithm: assign the image with the highest similarity ▪ Detracting from merit of mosaic art ▪ Same creation method for all use cases Phosaiq: consider as a combinatorial optimization problem ▪ Create mosaic art with merit of mosaic ▪ Deal with various use cases ▪ Consider data privacy
5. Formulation ▪ 5 Assign an image with higher similarity Consider the relations between two sections or images Assign only one image to each section
6. Example 1: Logo of the contest ▪ Materials are slides of participating teams ▪ Consider assigning images and timeline 6
7. Example 2: Flag of Canada ▪ Red maple and snow in Canada ▪ Penalty for assigning same photo in adjacent section 7
8. Use case 1: Artwork Royal Space Force – The Wings of Honneamise A legendary animation in Japan 35 years ago 8 ANIME MUSEUM Vol.01 (https://animuse.jp)
9. Use case 2: Souvenir (1) ▪ Equalize the number of times each person is in the picture ▪ Test: graduation from Nogizaka46 (Japanese idol group) 9 Large bias Small bias Without Equalization With Equalization
10. Use case 2: Souvenir (2) ▪ Graduation of preschoolers (including our co-worker's child) ▪ Give priority to images with: ▫ larger faces (by face detection) ▫ faces facing forward (by head pose estimation) 10
11. System design for privacy Just send similarities, not image data 11 (2) Similarity (4) QUBO model (6) Solutions (7) Arrangement of images (1) Client: Calculate the similarity of images (3) Server: Generate QUBO model (5) Quantum Annealer: Optimize the arrangement of images
12. Conclusion ▪ Optimization technique can create photo mosaic: ▫ with mosaic goodness ▫ suitable for various use cases ▪ Quantum Annealer (and Machine Learning) expand the potential ▪ System design to protect user’s privacy ▪ Exploring different representations, not just images ▪ Exploring new values of photo mosaic (to be continued…) ▫ Bring awareness to users ▫ Encourage them to take action 12
13. THANKS! You can find us at: phosaiq@gmail.com https:/ /github.com/phosaiq 13