image_id int64 0 519 | query_object int64 1 11 | occlusion_paths listlengths 1 6 | top_objects listlengths 1 4 | depends_on listlengths 0 4 | object_semantic_id int64 1 88 | ambiguity stringclasses 1
value | layer int64 1 7 | difficulty stringclasses 3
values | k_min int64 0 6 | num_paths int64 0 6 | new_difficulty stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 1 | [
[
1
]
] | [
1
] | [] | 16 | 1 | Easy | 0 | 0 | ||
1 | 1 | [
[
1
]
] | [
1
] | [] | 39 | 1 | Easy | 0 | 0 | ||
2 | 1 | [
[
1
]
] | [
1
] | [] | 25 | 1 | Easy | 0 | 0 | ||
3 | 1 | [
[
1
]
] | [
1
] | [] | 11 | 1 | Easy | 0 | 0 | ||
4 | 1 | [
[
1
]
] | [
1
] | [] | 69 | 1 | Easy | 0 | 0 | ||
5 | 1 | [
[
1
]
] | [
1
] | [] | 16 | 1 | Easy | 0 | 0 | ||
6 | 1 | [
[
1
]
] | [
1
] | [] | 51 | 1 | Easy | 0 | 0 | ||
7 | 1 | [
[
1
]
] | [
1
] | [] | 46 | 1 | Easy | 0 | 0 | ||
8 | 1 | [
[
1
]
] | [
1
] | [] | 3 | 1 | Easy | 0 | 0 | ||
9 | 1 | [
[
1
]
] | [
1
] | [] | 28 | 1 | Easy | 0 | 0 | ||
10 | 1 | [
[
1
]
] | [
1
] | [] | 17 | 1 | Easy | 0 | 0 | ||
11 | 1 | [
[
1
]
] | [
1
] | [] | 47 | 1 | Easy | 0 | 0 | ||
12 | 1 | [
[
1
]
] | [
1
] | [] | 63 | 1 | Easy | 0 | 0 | ||
13 | 1 | [
[
1
]
] | [
1
] | [] | 60 | 1 | Easy | 0 | 0 | ||
14 | 1 | [
[
1
]
] | [
1
] | [] | 53 | 1 | Easy | 0 | 0 | ||
15 | 1 | [
[
1
]
] | [
1
] | [] | 8 | 1 | Easy | 0 | 0 | ||
16 | 1 | [
[
1
]
] | [
1
] | [] | 58 | 1 | Easy | 0 | 0 | ||
17 | 1 | [
[
1
]
] | [
1
] | [] | 18 | 1 | Easy | 0 | 0 | ||
18 | 1 | [
[
1
]
] | [
1
] | [] | 27 | 1 | Easy | 0 | 0 | ||
19 | 1 | [
[
1
]
] | [
1
] | [] | 27 | 1 | Easy | 0 | 0 | ||
20 | 1 | [
[
1
]
] | [
1
] | [] | 43 | 1 | Easy | 0 | 0 | ||
21 | 1 | [
[
1
]
] | [
1
] | [] | 11 | 1 | Easy | 0 | 0 | ||
22 | 1 | [
[
1
]
] | [
1
] | [] | 18 | 1 | Easy | 0 | 0 | ||
23 | 1 | [
[
1
]
] | [
1
] | [] | 28 | 1 | Easy | 0 | 0 | ||
24 | 1 | [
[
1
]
] | [
1
] | [] | 62 | 1 | Easy | 0 | 0 | ||
25 | 1 | [
[
1
]
] | [
1
] | [] | 48 | 1 | Easy | 0 | 0 | ||
26 | 1 | [
[
1
]
] | [
1
] | [] | 11 | 1 | Easy | 0 | 0 | ||
27 | 1 | [
[
1
]
] | [
1
] | [] | 13 | 1 | Easy | 0 | 0 | ||
28 | 1 | [
[
1
]
] | [
1
] | [] | 47 | 1 | Easy | 0 | 0 | ||
29 | 1 | [
[
1
]
] | [
1
] | [] | 18 | 1 | Easy | 0 | 0 | ||
30 | 1 | [
[
1
]
] | [
1
] | [] | 3 | 1 | Easy | 0 | 0 | ||
31 | 1 | [
[
1
]
] | [
1
] | [] | 62 | 1 | Easy | 0 | 0 | ||
32 | 1 | [
[
1
]
] | [
1
] | [] | 61 | 1 | Easy | 0 | 0 | ||
33 | 1 | [
[
1
]
] | [
1
] | [] | 24 | 1 | Easy | 0 | 0 | ||
34 | 1 | [
[
1
]
] | [
1
] | [] | 44 | 1 | Easy | 0 | 0 | ||
35 | 1 | [
[
1
]
] | [
1
] | [] | 24 | 1 | Easy | 0 | 0 | ||
36 | 1 | [
[
1
]
] | [
1
] | [] | 16 | 1 | Easy | 0 | 0 | ||
37 | 1 | [
[
1
]
] | [
1
] | [] | 65 | 1 | Easy | 0 | 0 | ||
38 | 1 | [
[
1
]
] | [
1
] | [] | 3 | 1 | Easy | 0 | 0 | ||
39 | 1 | [
[
1
]
] | [
1
] | [] | 47 | 1 | Easy | 0 | 0 | ||
40 | 1 | [
[
1
]
] | [
1
] | [] | 45 | 1 | Easy | 0 | 0 | ||
41 | 1 | [
[
1
]
] | [
1
] | [] | 7 | 1 | Easy | 0 | 0 | ||
42 | 1 | [
[
1
]
] | [
1
] | [] | 7 | 1 | Easy | 0 | 0 | ||
43 | 1 | [
[
1
]
] | [
1
] | [] | 17 | 1 | Easy | 0 | 0 | ||
44 | 1 | [
[
1
]
] | [
1
] | [] | 52 | 1 | Easy | 0 | 0 | ||
45 | 1 | [
[
1
]
] | [
1
] | [] | 65 | 1 | Easy | 0 | 0 | ||
46 | 1 | [
[
1
]
] | [
1
] | [] | 27 | 1 | Easy | 0 | 0 | ||
47 | 1 | [
[
1
]
] | [
1
] | [] | 51 | 1 | Easy | 0 | 0 | ||
48 | 1 | [
[
1
]
] | [
1
] | [] | 54 | 1 | Easy | 0 | 0 | ||
49 | 1 | [
[
1
]
] | [
1
] | [] | 16 | 1 | Easy | 0 | 0 | ||
50 | 1 | [
[
1
]
] | [
1
] | [] | 51 | 1 | Easy | 0 | 0 | ||
51 | 1 | [
[
1
]
] | [
1
] | [] | 60 | 1 | Easy | 0 | 0 | ||
52 | 1 | [
[
1
]
] | [
1
] | [] | 44 | 1 | Easy | 0 | 0 | ||
53 | 1 | [
[
1
]
] | [
1
] | [] | 7 | 1 | Easy | 0 | 0 | ||
54 | 1 | [
[
1
]
] | [
1
] | [] | 12 | 1 | Easy | 0 | 0 | ||
55 | 1 | [
[
1
]
] | [
1
] | [] | 51 | 1 | Easy | 0 | 0 | ||
56 | 1 | [
[
1
]
] | [
1
] | [] | 53 | 1 | Easy | 0 | 0 | ||
57 | 1 | [
[
1
]
] | [
1
] | [] | 12 | 1 | Easy | 0 | 0 | ||
57 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 59 | 2 | Medium | 1 | 1 | ||
58 | 1 | [
[
1
]
] | [
1
] | [] | 15 | 1 | Easy | 0 | 0 | ||
58 | 3 | [
[
2,
3
]
] | [
2
] | [
2
] | 18 | 2 | Medium | 1 | 1 | ||
59 | 1 | [
[
1
]
] | [
1
] | [] | 13 | 1 | Easy | 0 | 0 | ||
59 | 4 | [
[
3,
4
]
] | [
3
] | [
3
] | 61 | 2 | Medium | 1 | 1 | ||
60 | 1 | [
[
1
]
] | [
1
] | [] | 12 | 1 | Easy | 0 | 0 | ||
60 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 49 | 2 | Medium | 1 | 1 | ||
61 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 48 | 2 | Medium | 1 | 1 | ||
62 | 1 | [
[
1
]
] | [
1
] | [] | 18 | 1 | Easy | 0 | 0 | ||
62 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 28 | 2 | Medium | 1 | 1 | ||
63 | 2 | [
[
3,
2
]
] | [
3
] | [
3
] | 12 | 2 | Medium | 1 | 1 | ||
64 | 1 | [
[
1
]
] | [
1
] | [] | 65 | 1 | Easy | 0 | 0 | ||
64 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 62 | 2 | Medium | 1 | 1 | ||
65 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 28 | 2 | Medium | 1 | 1 | ||
65 | 2 | [
[
2
]
] | [
2
] | [] | 49 | 1 | Easy | 0 | 0 | ||
66 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 22 | 2 | Medium | 1 | 1 | ||
66 | 2 | [
[
2
]
] | [
2
] | [] | 66 | 1 | Easy | 0 | 0 | ||
67 | 3 | [
[
1,
3
]
] | [
1
] | [
1
] | 58 | 2 | Medium | 1 | 1 | ||
68 | 1 | [
[
1
]
] | [
1
] | [] | 52 | 1 | Easy | 0 | 0 | ||
69 | 4 | [
[
3,
4
]
] | [
3
] | [
3
] | 30 | 2 | Medium | 1 | 1 | ||
70 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 43 | 2 | Medium | 1 | 1 | ||
71 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 59 | 2 | Medium | 1 | 1 | ||
72 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 29 | 2 | Medium | 1 | 1 | ||
72 | 2 | [
[
2
]
] | [
2
] | [] | 15 | 1 | Easy | 0 | 0 | ||
73 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 28 | 2 | Medium | 1 | 1 | ||
73 | 2 | [
[
2
]
] | [
2
] | [] | 40 | 1 | Easy | 0 | 0 | ||
74 | 1 | [
[
1
]
] | [
1
] | [] | 16 | 1 | Easy | 0 | 0 | ||
74 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 56 | 2 | Medium | 1 | 1 | ||
75 | 1 | [
[
1
]
] | [
1
] | [] | 11 | 1 | Easy | 0 | 0 | ||
76 | 1 | [
[
2,
1
]
] | [
2
] | [
2
] | 63 | 2 | Medium | 1 | 1 | ||
76 | 2 | [
[
2
]
] | [
2
] | [] | 61 | 1 | Easy | 0 | 0 | ||
77 | 4 | [
[
1,
4
]
] | [
1
] | [
1
] | 27 | 2 | Medium | 1 | 1 | ||
78 | 1 | [
[
1
]
] | [
1
] | [] | 18 | 1 | Easy | 0 | 0 | ||
78 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 57 | 2 | Medium | 1 | 1 | ||
79 | 1 | [
[
1
]
] | [
1
] | [] | 12 | 1 | Easy | 0 | 0 | ||
80 | 1 | [
[
1
]
] | [
1
] | [] | 53 | 1 | Easy | 0 | 0 | ||
80 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 49 | 2 | Medium | 1 | 1 | ||
81 | 1 | [
[
1
]
] | [
1
] | [] | 64 | 1 | Easy | 0 | 0 | ||
81 | 2 | [
[
1,
2
]
] | [
1
] | [
1
] | 62 | 2 | Medium | 1 | 1 | ||
82 | 1 | [
[
1
]
] | [
1
] | [] | 12 | 1 | Easy | 0 | 0 | ||
82 | 4 | [
[
2,
4
]
] | [
2
] | [
2
] | 53 | 2 | Medium | 1 | 1 | ||
83 | 3 | [
[
4,
3
]
] | [
4
] | [
4
] | 63 | 2 | Medium | 1 | 1 |
VOCC-Grasp — MetaGraspNet-V2 real evaluation subset
The real-world data needed to run VOCC-Grasp on MetaGraspNet-V2. This is not the full real capture set (818 scenes, 4 views each, 30 GB); it is the filtered evaluation subset — 838 cases over 511 images — with RGB, depth, real intrinsics and instance masks for every one of them. No file is missing.
The synthetic half of the benchmark is chiencn/vocc_synthetic.
Contents
| file | size | contents |
|---|---|---|
scenes.tar.gz |
3.83 GB | 511 raw captures: 3.npz (depth + instances_objects), 3_rgb.png (1944×1200), 3_camera_params.json |
images.tar.gz |
987.0 MB | 511 × images/image_%06d.png, 1200×1200 RGB |
masks_crop.tar.gz |
5.0 MB | 511 × masks_npy_real_crop/image_%06d.npy, int32 1200×1200 |
masks_full.tar.gz |
7.0 MB | 511 × masks_npy_real/image_%06d.npy, int32 1200×1944 — UnoGrasp baseline only |
meta/ |
~1.8 MB | id mapping, object names, ground truths, NLP prompts |
Extracted footprint is about 13 GB — the mask tarballs are tiny because raw int32 instance maps gzip roughly 350×.
meta/
| file | contents |
|---|---|
real_world_mapping_fixed.json |
"%06d" image id → sceneN_viewM. Required |
real_object_names.json |
image id → object id → name, for the VLM prompt. Required |
test_GT_subset_hardall_easy300_medium300.json |
the 838-case split: 300 Easy / 300 Medium / 238 Hard (also in the git repository) |
gt_from_occlusion_real_v1.json |
the wider 1480-case ground truth over 802 scenes — see the caveat below |
nlp_subset_hardall_e300_m300.jsonl |
the same 838 cases as conversation prompts, for the UnoGrasp baseline |
manifest.sha256 |
checksums of the four tarballs |
Geometry
Every one of the 520 mapping entries uses view 3, so exactly one view per scene ships here.
Depth is in centimetres, float32, and about 20% of it is NaN — real sensor dropout, not a
packaging artefact. Zero it before back-projecting.
Unlike the synthetic set, these carry real intrinsics in 3_camera_params.json
(fx, fy, cx, cy, width=1944, height=1200). The 1200-wide benchmark files are the centred crop
raw[:, 372:1572] of the 1944-wide captures — byte-identical, verified — so a cropped frame
keeps fx, fy, cy and takes cx - 372. grasp_viz/real_data.py in the repository does this.
Instance maps are integer object ids; 0 is background. masks_npy_real_crop/ is
3.npz['instances_objects'] cropped and cast to int32, so it is redundant with scenes.tar.gz —
it ships because the pipeline reads it directly and it costs 5 MB.
Usage
pip install huggingface_hub
hf download chiencn/vocc_real --repo-type dataset --local-dir /tmp/real_dl
cd /path/to/vocc-grasp
for f in scenes images masks_crop masks_full; do
tar -xzf /tmp/real_dl/$f.tar.gz -C .
done
cp /tmp/real_dl/meta/real_world_mapping_fixed.json \
/tmp/real_dl/meta/real_object_names.json \
/tmp/real_dl/meta/gt_from_occlusion_real_v1.json .
Everything extracts at the repository root, which is the layout the code expects:
vocc-grasp
├── real_world_mapping_fixed.json
├── real_object_names.json
├── images/image_000000.png
├── masks_npy_real_crop/image_000000.npy
├── masks_npy_real/image_000000.npy
└── data_ifl_0/mnt/data1/data_ifl_real/scene0
├── 3.npz
├── 3_rgb.png
└── 3_camera_params.json
The data_ifl_<N> shard names are the upstream ones and are preserved; the loader globs
data_ifl_*/mnt/data1/data_ifl_real/scene*, so the split across shards does not matter.
Caveat on gt_from_occlusion_real_v1.json
run_uoais_pipeline_real.py takes that file as its default --gt-path, but it spans 1480 cases
over 802 scenes while only the 511 evaluated scenes are published here. Left at the default it
will skip the ~291 scenes it cannot find. Pass the subset ground truth instead:
python run_uoais_pipeline_real.py --gt-path test_GT_subset_hardall_easy300_medium300.json
Scope
Only the 511 scenes the evaluation split touches are packaged, view 3 only. Not included: the
other 307 scenes, the other three views, and the per-scene amodal / grasps / order /
bin_rel / scene.hdf5 extras — no part of the VOCC-Grasp pipeline reads them.
Reproducing the reported real score needs no download at all; the fused edge scores and the ground truth are both in the git repository. This dataset is for regenerating predictions from raw RGB-D, and for running the grasp stage on real scenes.
Licence and attribution
Derived from MetaGraspNet-V2; this redistribution follows its terms and is for academic use only. Cite the upstream dataset when you use it.
- Downloads last month
- 37