Datasets:

bhatvineet
/

mg-vqa-bench

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema_version: int64
scene_id: string
episode: int64
episode_seed: int64
generation_seed: int64
replay_seed: int64
question_type: string
benchmark_version: string
asset_path: string
img_size: int64
n_objects: int64
table_texture: struct<source_name: string, store_rel: string, hash: string, texture_seed: int64, files: list<item:  (... 38 chars omitted)
  child 0, source_name: string
  child 1, store_rel: string
  child 2, hash: string
  child 3, texture_seed: int64
  child 4, files: list<item: struct<path: string, sha256: string>>
      child 0, item: struct<path: string, sha256: string>
          child 0, path: string
          child 1, sha256: string
table: struct<texture_seed: int64, texture_dir_name: string, texture_hash: string, texture_store_rel: strin (... 24 chars omitted)
  child 0, texture_seed: int64
  child 1, texture_dir_name: string
  child 2, texture_hash: string
  child 3, texture_store_rel: string
  child 4, texture_copied: null
objects: list<item: struct<object_key: string, spawn_index: int64, obj_id: int64, name: string, folder: strin (... 379 chars omitted)
  child 0, item: struct<object_key: string, spawn_index: int64, obj_id: int64, name: string, folder: string, base_pos (... 367 chars omitted)
      child 0, object_key: string
      child 1, spawn_index: int64
      child 2, obj_id: int64
      child 3, name: string
      child 4, folder: string
      child 5, base_position: list<item: double>
          child 0, item: double
      child 6, base_orientat
...
wer_text: null, question_descriptor: struct<q (... 426 chars omitted)
  child 0, schema_version: int64
  child 1, gt_answer: string
  child 2, gt_answer_text: null
  child 3, question_descriptor: struct<question: string, question_type: string, question_subtype: string, gt_answer: string, answer_ (... 292 chars omitted)
      child 0, question: string
      child 1, question_type: string
      child 2, question_subtype: string
      child 3, gt_answer: string
      child 4, answer_options: list<item: string>
          child 0, item: string
      child 5, target_name: string
      child 6, cover_object: string
      child 7, also_valid: list<item: string>
          child 0, item: string
      child 8, protected_obj_ids: list<item: int64>
          child 0, item: int64
      child 9, require_target_visible: bool
      child 10, reveal_area: int64
      child 11, question_stem: string
      child 12, phrasing_source: string
      child 13, template_variant: string
      child 14, template_category: string
  child 4, gt_edited: bool
  child 5, gt_edit_reason: string
by_type: struct<beneath: int64, compare: int64, count: int64, find: int64>
  child 0, beneath: int64
  child 1, compare: int64
  child 2, count: int64
  child 3, find: int64
description: string
episodes: list<item: struct<ep_dir: string, question_type: string>>
  child 0, item: struct<ep_dir: string, question_type: string>
      child 0, ep_dir: string
      child 1, question_type: string
n_episodes: int64
name: string
to
{'schema_version': Value('int64'), 'name': Value('string'), 'description': Value('string'), 'n_episodes': Value('int64'), 'by_type': {'beneath': Value('int64'), 'compare': Value('int64'), 'count': Value('int64'), 'find': Value('int64')}, 'episodes': List({'ep_dir': Value('string'), 'question_type': Value('string')})}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              schema_version: int64
              scene_id: string
              episode: int64
              episode_seed: int64
              generation_seed: int64
              replay_seed: int64
              question_type: string
              benchmark_version: string
              asset_path: string
              img_size: int64
              n_objects: int64
              table_texture: struct<source_name: string, store_rel: string, hash: string, texture_seed: int64, files: list<item:  (... 38 chars omitted)
                child 0, source_name: string
                child 1, store_rel: string
                child 2, hash: string
                child 3, texture_seed: int64
                child 4, files: list<item: struct<path: string, sha256: string>>
                    child 0, item: struct<path: string, sha256: string>
                        child 0, path: string
                        child 1, sha256: string
              table: struct<texture_seed: int64, texture_dir_name: string, texture_hash: string, texture_store_rel: strin (... 24 chars omitted)
                child 0, texture_seed: int64
                child 1, texture_dir_name: string
                child 2, texture_hash: string
                child 3, texture_store_rel: string
                child 4, texture_copied: null
              objects: list<item: struct<object_key: string, spawn_index: int64, obj_id: int64, name: string, folder: strin (... 379 chars omitted)
                child 0, item: struct<object_key: string, spawn_index: int64, obj_id: int64, name: string, folder: string, base_pos (... 367 chars omitted)
                    child 0, object_key: string
                    child 1, spawn_index: int64
                    child 2, obj_id: int64
                    child 3, name: string
                    child 4, folder: string
                    child 5, base_position: list<item: double>
                        child 0, item: double
                    child 6, base_orientat
              ...
              wer_text: null, question_descriptor: struct<q (... 426 chars omitted)
                child 0, schema_version: int64
                child 1, gt_answer: string
                child 2, gt_answer_text: null
                child 3, question_descriptor: struct<question: string, question_type: string, question_subtype: string, gt_answer: string, answer_ (... 292 chars omitted)
                    child 0, question: string
                    child 1, question_type: string
                    child 2, question_subtype: string
                    child 3, gt_answer: string
                    child 4, answer_options: list<item: string>
                        child 0, item: string
                    child 5, target_name: string
                    child 6, cover_object: string
                    child 7, also_valid: list<item: string>
                        child 0, item: string
                    child 8, protected_obj_ids: list<item: int64>
                        child 0, item: int64
                    child 9, require_target_visible: bool
                    child 10, reveal_area: int64
                    child 11, question_stem: string
                    child 12, phrasing_source: string
                    child 13, template_variant: string
                    child 14, template_category: string
                child 4, gt_edited: bool
                child 5, gt_edit_reason: string
              by_type: struct<beneath: int64, compare: int64, count: int64, find: int64>
                child 0, beneath: int64
                child 1, compare: int64
                child 2, count: int64
                child 3, find: int64
              description: string
              episodes: list<item: struct<ep_dir: string, question_type: string>>
                child 0, item: struct<ep_dir: string, question_type: string>
                    child 0, ep_dir: string
                    child 1, question_type: string
              n_episodes: int64
              name: string
              to
              {'schema_version': Value('int64'), 'name': Value('string'), 'description': Value('string'), 'n_episodes': Value('int64'), 'by_type': {'beneath': Value('int64'), 'compare': Value('int64'), 'count': Value('int64'), 'find': Value('int64')}, 'episodes': List({'ep_dir': Value('string'), 'question_type': Value('string')})}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

mg-vqa-bench

MG-VQA-Bench contains 600 frozen visual question answering tasks for MG-VQA-Sim. A vision-language model answers questions about cluttered tabletop scenes using a single image, perception tools, or perception and robot manipulation.

Task	Questions
Beneath	150
Compare	150
Count	150
Find	150

The scenes use 319 unique object assets. Each snapshot stores object poses and physics parameters, the question, and the annotations needed for evaluation. Compare scenes include baked A/B/C badge textures. Shared table textures are stored once and referenced by relative path.

Download

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="bhatvineet/mg-vqa-bench",
    repo_type="dataset",
    local_dir="mg-vqa-bench",
)

For reproducible experiments, set revision to a commit SHA from this repository.

Run

Install MG-VQA-Sim and its shared object assets before evaluation. Object meshes, object URDFs, robot assets, and model weights are separate from this dataset. Even Direct mode requires the simulator and object assets to render the scenes.

From the MG-VQA-Sim source directory:

python scripts/download_assets.py
./run_mgvqa.sh --dataset /path/to/mg-vqa-bench \
  --model azure/openai/gpt-6-astra --mode agentic --workers 4

Use --mode direct, --mode perception, or --mode agentic. All 600 episodes are evaluated by default. Accuracy is correct answers divided by answered questions; non-answers are reported separately.

Layout

bank_manifest.json                 # Episode order and task types
_texture_store/<hash>/             # Shared table texture maps
ep_NNN_<task>/scene_snapshot/
  scene_snapshot.json              # Frozen scene, question and scoring annotation
  badges/                          # Compare badge textures, where applicable

The authoritative evaluation question and answer are in annotation.question_descriptor inside each snapshot. This annotation is for the evaluator, not the model prompt. Retain the directory structure so texture references resolve correctly. Shared object folders are resolved against the MG-VQA-Sim asset installation.

Asset sources and terms

Assets and their derivatives retain their respective upstream licenses. Refer to the original sources for terms of use:

Object source	Objects	Upstream source and terms
Meta Digital Twin Catalog	136	Source · Terms
MegaPose-sourced object models	155	Data sources
BOP: HB, HANDAL, HOPE, YCB-V, XYZ-IBD, ITODD and LM-O	28	Datasets and terms

Table textures: ambientCG. The MG-VQA source-code license does not relicense third-party assets or their derivatives.

Downloads last month: 494

Total file size:

2.33 GB