This README contains figures for the above paper.
Meta-learned initialization + 4 gradient steps and target for test scene.
Course of optimization for imputation of voxel from partial observation.
Partial observation | Imputation |
---|---|
![]() |
![]() |
Partial observation | Imputation |
---|---|
![]() |
![]() |
Partial observation | Imputation |
---|---|
![]() |
![]() |
Partial observation | Imputation |
---|---|
![]() |
![]() |
Uncurated samples from DDPM (diffusion) trained on 64-dim modulations of SRN-cars.
Latent interpolation between two car scenes with moving pose.
Novel view synthesis from occluded view.
Occluded view | Ground truth | Inferred | No prior |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Uncurated samples from flow trained on 256-dim modulations on ERA-5 temperature data.
Additional voxel imputation results.
Partial observation | Imputation |
---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Additional novel view synthesis results.
Occluded view | Ground truth | Inferred | No prior |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |