GraspNet-Infinite object splits: test_easy, test_hard, test_medium, train_4, train_5, train_6

README.md

@@ -18,9 +18,9 @@ Textured, closed, single-object 3D models for tabletop grasping scenes, packed b
 | `train_1.zip` | 16 | 117 MB |
 | `train_2.zip` | 16 | 85 MB |
 | `train_3.zip` | 32 | 145 MB |
-| `train_4.zip` | 63 | 275 MB |
-| `train_5.zip` | 126 | 601 MB |
-| `train_6.zip` | 252 | 972 MB |
+| `train_4.zip` | 63 | 272 MB |
+| `train_5.zip` | 126 | 592 MB |
+| `train_6.zip` | 252 | 962 MB |
 | `train_7.zip` | 496 | 2053 MB |
 | `train_8.zip` | 495 | 2100 MB |
 | `train_9.zip` | 500 | 2047 MB |
@@ -28,10 +28,10 @@ Textured, closed, single-object 3D models for tabletop grasping scenes, packed b
 | `train_11.zip` | 494 | 1658 MB |
 | `train_12.zip` | 491 | 1910 MB |
 | `test_easy.zip` | 99 | 296 MB |
-| `test_medium.zip` | 99 | 399 MB |
-| `test_hard.zip` | 93 | 392 MB |
+| `test_medium.zip` | 99 | 396 MB |
+| `test_hard.zip` | 93 | 337 MB |
 
-`train_1` to `train_12` are disjoint random groups of the training objects. Groups 1 to 9 (16, 16, 32, 64, 128, 256, 512, 512, 512 objects) are drawn from the objects with a real image texture, and groups 1 to 3 passed a visual quality check. Groups 10 to 12 (510, 510 and 509 objects) hold every other training object: image-textured objects not drawn into groups 1 to 9, flat-shaded multi-coloured models, and objects judged textured on sight. `test_easy`, `test_medium` and `test_hard` are the three 100-object test sets, chosen by category and confirmed visually: easy = primitive shapes, medium = geometry composites (hammers, mugs, bowls, toys), hard = very complex geometry (dragons, birds with spread wings, figurines). Training and test objects never share an object.
+`train_1` to `train_12` are disjoint random groups of the training objects. Groups 1 to 9 (sizes in the table above; 16, 16, 32, 64, 128, 256 and about 500 each for groups 7 to 9) are drawn from the objects with a real image texture, and groups 1 to 3 passed a visual quality check. Groups 10 to 12 (about 490 objects each, sizes in the table above) hold every other training object: image-textured objects not drawn into groups 1 to 9, flat-shaded multi-coloured models, and objects judged textured on sight. `test_easy`, `test_medium` and `test_hard` are the three 100-object test sets, chosen by category and confirmed visually: easy = primitive shapes, medium = geometry composites (hammers, mugs, bowls, toys), hard = very complex geometry (dragons, birds with spread wings, figurines). Training and test objects never share an object.
 
 Each zip holds one folder per object: `<uid>/<uid>.glb` (Objaverse, textures embedded) or `<uid>/model.obj` + `model.mtl` + `materials/textures/*.png` (Google Scanned Objects), and `<uid>/<uid>.json` with identity, sub-category, family, assigned longest extent in cm, the scale factor to metres, licence and attribution. `manifest.csv` at the root lists every object with its group; `attribution.csv` lists title, author, URL and licence for CC-BY attribution.
 

test_easy.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ac8b4794bf4c2c7ab7b31fddb6338bd6a87828634a3a2ae831c2b5061d857eb
-size 296430389
+oid sha256:9b955ba76ee9fd76cf09e07dae57b69b69c5493ffdc95497694de384f4178190
+size 295546061

test_hard.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4f4d2db7b1b69a88a33df32299a13e9c5149efbedab91562c96001723c34780
-size 391925919
+oid sha256:d370e3aab7c450dd6b43f2f7d4229186af77d8a3b9b4fc9534d1d53011e8861a
+size 337193583

test_medium.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cfd1158716f69241bba809e888fb59cc698f66f6959e2868b5338d70c5ad4a28
-size 398866195
+oid sha256:4f15a8e9ef16f43a008cc49672535a4241e64d2a78a700bb5abca5d17d440138
+size 396124193

train_4.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef5d3988c331dca32b38a04262f9af98077e2b24768a05afe3aa7feb5dae18df
-size 274653285
+oid sha256:04cfd26757c15971084f59522470af36d037c398f03f29dd978d545e5d9641ac
+size 271863901

train_5.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09d5f2db2c67a65f2aca5df5d1603bb0e67be7f1bd94277f1e5e634a8ffeb1e9
-size 601366619
+oid sha256:fc9d32dd60d8a22ed6695d64bfecf43709d4f0fdb7b22a5251012078b2cfa723
+size 592111262

train_6.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a71580f8a42aca65b9f53d1dba5c67bd0b1568dab4f1e4de6e6d42a40c5c15e1
-size 971567438
+oid sha256:d779dbfdd2ba6852e298eaaa41288d5fc0d132fcee2d3b8f26b2b28a60f19233
+size 962431449