Scene and View
- Scene:
- A single arrangement of objects in the bin.
- Different environments.
- Static Vs. Falling Objects
- View:
- Static 9 point of views (45' + top view)
-
Random point of views
-
Note: There is no relation between # objects and (# Scene and #Views).
DATASET Scenarios
- a) level 1: no occlusion.
- b) level 2: some occlusion.
- c) level 3: incomplete objects (scissors is crosscut by the banana).
- d) level 4: multiple instances of the same object class.
- e) level 5: includes all difficult characteristics (incomplete and non-unique objects).
Formula
- #images = #scenes * #view-point
- size of dataset = #images * 42 MB
| Version | #Images | #Scenes | #Views | Size | Splitting (Tr, Vl, Ts) |
|---|---|---|---|---|---|
| 1 | 20K (LineMOD) | 15 | 1200 | 800 GB | |
| 2 | ~60k (FAT) | 15 (3S * 5L) | - | 2.5 TB | |
| 3 | 80k (T-LESS) | 20 | - | 3.2 TB | |
| 4 | 97K (Grasp-NET) | 190 | - | 4.1 TB | |
| 5 | 134K (YCB-V) | 92 | - | 5.6 TB | |
| 6 | 206K (BIN-P) | - | - | 8.6 TB | |
| 7 | 396K (StereOBJ-1M) | 183s , 11e | - | ~ 16 TB , X => (2t > 8.6 TB) | |
| 8 | 600K (ObjSynth) | 6 | 200 | ~ 42 TB , Y => (3t > 8.6 TB) | |
| 9 (X10) | 6M (Ours) | 6000 | 1000 | 10t > Y |
Splitting
- 15% 80% (LIneMod) same seq
- 80% 20% (YCB-V) diff seq
- 90% 10% (T-LESS)
- 50% 20% 30% (REAL)
- By Scenes {HOPE: (10, 10, 40)}
- Randomly (DOPose)