saar stern

[13/n] PRISM achieves the strongest correlation with humans in 3/4 categories — showing the power of combining diffusion priors with geometry-aware contrastive training.

[12/n] We evaluate correlation with human preferences across four axes: 🎯 Viewpoint Accuracy 🔁 Shared Consistency ✨ Plausibility of New Regions 📷 Image Quality Classic metrics (PSNR, SSIM, LPIPS, CLIP) and distribution scores (FID, FJD) fail on viewpoint-aware correctness.

🚀 Our paper “Appreciate the View: A Task-Aware Evaluation Framework for Novel View Synthesis” is accepted to 3DV 2026! @IdoSobol @orlitany 🔗 Project page: saarst.github.io/appreciate-the…