Representation Analysis for 3D Avatars 3B Avatarlar için Temsil Analizi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11637014
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: 3D avatar, CAP4D, cost-quality analysis, diffusion-based generation, Gaussian splatting
- Boğaziçi Üniversitesi Adresli: Evet
Özet
Recent generative models have enabled the creation of photorealistic 3D avatars from a limited number of images. However, the impact of generation settings on representation quality and computational cost has not been sufficiently studied. In this work, instead of proposing a new avatar generation method, we analyze the effect of different generation settings in a CAP4D-based 3D avatar pipeline. The results show that medium-cost configurations, where the generation time is reduced to approximately one quarter, can preserve visual quality close to high-cost settings. PSNR, LPIPS, CSIM and JOD results indicate that these configurations should be interpreted not as a strict quality improvement, but as an efficient cost-quality trade-off for practical use.