Fig. 7
From: Single view generalizable 3D reconstruction based on 3D Gaussian splatting
Qualitative comparisons on DTU, Real Forward-Facing, and Tanks and Temples datasets. In each pair, the left image is the ground truth and the right image is the rendering result from our model finetuned on the DTU dataset. Red boxes highlight areas with noticeable reconstruction details and differences. Despite being trained only on DTU, our method shows strong generalization across domains, producing visually faithful results even on unseen datasets.
