Release NeRF Synthetic Blender processed data

DATA_MANIFEST.json

@@ -2,7 +2,7 @@
   "release": "AnchorSplat processed third-party datasets",
   "date": "2026-07-03",
   "layout": "Extract archives into the AnchorSplat repository data/ directory.",
-  "note": "The MVImgNet rebuttal numbers use mvimgnet_supergaussian_4x_downsampled_lr.",
+  "note": "MVImgNet rebuttal numbers use mvimgnet_supergaussian_4x_downsampled_lr. NeRF Synthetic Blender sparse-38 bicubic is included for zero-shot evaluation.",
   "datasets": [
     {
       "name": "mvimgnet_supergaussian_4x_downsampled_lr",
@@ -84,6 +84,43 @@
         "0__3c005f04"
       ],
       "missing_required_files": {}
+    },
+    {
+      "name": "nerf_synthetic_sparse38_bicubic",
+      "description": "Processed NeRF Synthetic Blender sparse-38 bicubic split with 200-resolution input 3DGS and 800-resolution target views.",
+      "archive": "nerf_synthetic_sparse38_bicubic.tar.zst",
+      "sha256": "2e2fa15c782b54d5cef75fbea9db82fcd2e234e17e64ce0b50553951dea3677e",
+      "anchor_splat_config": "configs/dataset/nerf_synthetic_sparse38_bicubic.gin",
+      "scene_count": 8,
+      "file_count": 672,
+      "scenes": [
+        "chair",
+        "drums",
+        "ficus",
+        "hotdog",
+        "lego",
+        "materials",
+        "mic",
+        "ship"
+      ],
+      "missing_required_files": {},
+      "metrics": {
+        "input": {
+          "psnr": 28.9134,
+          "ssim": 0.9166,
+          "lpips": 0.1544
+        },
+        "anchorsplat_1x": {
+          "psnr": 26.902,
+          "ssim": 0.9006,
+          "lpips": 0.1302
+        },
+        "anchorsplat_20x": {
+          "psnr": 26.9557,
+          "ssim": 0.9022,
+          "lpips": 0.1131
+        }
+      }
     }
   ]
 }

README.md

@@ -7,8 +7,13 @@ pretty_name: AnchorSplat Processed Third-Party Data
 
 # AnchorSplat Processed Third-Party Data
 
-Processed third-party MVImgNet assets for AnchorSplat evaluation.
+Processed third-party evaluation assets for AnchorSplat.
 
-The ECCV 2026 rebuttal MVImgNet zero-shot table uses `mvimgnet_supergaussian_4x_downsampled_lr` with `configs/dataset/mvimgnet_supergaussian_4x_downsampled_lr.gin`.
+Included packages:
+
+- `mvimgnet_supergaussian_4x_downsampled_lr`: MVImgNet split used for the ECCV 2026 rebuttal zero-shot table.
+- `mvimgnet_supergaussian`: MVImgNet processed from SuperGaussian xyz/rgb inputs.
+- `mvimgnet_lowres_3dgs`: MVImgNet processed with low-resolution 3DGS inputs.
+- `nerf_synthetic_sparse38_bicubic`: NeRF Synthetic Blender sparse-38 bicubic split with 200-resolution input 3DGS and 800-resolution target views.
 
 Extract archives into the AnchorSplat repository `data/` directory. See `DATA_MANIFEST.json` and `SHA256SUMS.txt` for scene lists and checksums.

SHA256SUMS.txt

@@ -1,3 +1,4 @@
+9dd15a50e3b46156b07dcb220e9f2523c89655fd21ceb6dc8247799fdbdb2958  mvimgnet_lowres_3dgs.tar.zst
 66c6b43d99ce8bc8230eeca3ee9a61cb710597799b9afa9ae34f90e704bb8be0  mvimgnet_supergaussian_4x_downsampled_lr.tar.zst
 62b444328a898c990bf3a20d8c805a5e46335ae4b6532f54f11f53bb48847c92  mvimgnet_supergaussian.tar.zst
-9dd15a50e3b46156b07dcb220e9f2523c89655fd21ceb6dc8247799fdbdb2958  mvimgnet_lowres_3dgs.tar.zst
+2e2fa15c782b54d5cef75fbea9db82fcd2e234e17e64ce0b50553951dea3677e  nerf_synthetic_sparse38_bicubic.tar.zst

nerf_synthetic_sparse38_bicubic.tar.zst

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e2fa15c782b54d5cef75fbea9db82fcd2e234e17e64ce0b50553951dea3677e
+size 323413349