Datasets:

cp-bg-bench-anon
/

reproducibility

Error code:   StreamingRowsError
Exception:    TypeError
Message:      Couldn't cast array of type
struct<channelwise: bool, embed_dim: int64, freeze_backbone_when_no_lora: bool, hf_model: string, image_size: int64, in_channels: int64, lora: struct<alpha: int64, bias: string, dropout: double, enabled: bool, r: int64, targets: null, task_type: string>, name: string, native_channels: bool, projection_head: string>
to
{'embed_dim': Value('int64'), 'freeze_backbone_when_no_lora': Value('bool'), 'image_size': Value('int64'), 'in_channels': Value('int64'), 'lora': {'alpha': Value('int64'), 'bias': Value('string'), 'dropout': Value('float64'), 'enabled': Value('bool'), 'r': Value('int64'), 'targets': Value('null'), 'task_type': Value('string')}, 'name': Value('string'), 'native_channels': Value('bool'), 'normalizer': Value('string'), 'pool': Value('string'), 'pretrained': Value('bool'), 'projection_head': Value('string'), 'timm_name': Value('string')}
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                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 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 299, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2321, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2255, in cast_table_to_schema
                  cast_array_to_feature(
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 1804, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2101, in cast_array_to_feature
                  raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
              TypeError: Couldn't cast array of type
              struct<channelwise: bool, embed_dim: int64, freeze_backbone_when_no_lora: bool, hf_model: string, image_size: int64, in_channels: int64, lora: struct<alpha: int64, bias: string, dropout: double, enabled: bool, r: int64, targets: null, task_type: string>, name: string, native_channels: bool, projection_head: string>
              to
              {'embed_dim': Value('int64'), 'freeze_backbone_when_no_lora': Value('bool'), 'image_size': Value('int64'), 'in_channels': Value('int64'), 'lora': {'alpha': Value('int64'), 'bias': Value('string'), 'dropout': Value('float64'), 'enabled': Value('bool'), 'r': Value('int64'), 'targets': Value('null'), 'task_type': Value('string')}, 'name': Value('string'), 'native_channels': Value('bool'), 'normalizer': Value('string'), 'pool': Value('string'), 'pretrained': Value('bool'), 'projection_head': Value('string'), 'timm_name': Value('string')}

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.

cp-bg-bench — embeddings and inference checkpoints

Well- and cell-resolution embeddings, plus inference checkpoints, for the 36 model configurations evaluated in the cp-bg-bench benchmark (3 datasets × 3 image encoders × 4 input views).

Layout

<config_id>/
  aggregated.h5ad      # well-resolution: obs join keys + obsm["X_pca"], obsm["X_pca_harmony"]
  singlecell.h5ad      # per-cell embeddings produced by inference
  exports/
    imgenc_finetuned.json   # encoder config sidecar
    imgenc_finetuned.pth    # state_dict (preferred load path)
    imgenc_finetuned.pt     # pickled wrapper (legacy fallback)
manifest.json          # provenance: per-config file list and byte counts

<config_id> is <NN>_<dataset>_<imgenc>_<txtenc>_<view> where NN ∈ 01–36, dataset ∈ {JUMP, Rxrx1, Rxrx3C}, imgenc ∈ {DINO, OpenPhenom, SubCell}, txtenc ∈ {ECFP4, ESM2}, view ∈ {C, CD, S, SD}. See manifest.json for the exhaustive list.

Schema

Aggregated h5ad: obs carries (source, plate, well, batch, perturbation, is_control); obsm["X_pca"] and obsm["X_pca_harmony"] hold the 50-d PCA features used by every eval. Singlecell h5ad has the same schema at per-cell granularity.

Exports: imgenc_finetuned.json defines encoder_config (name, in_channels, image_size, embed_dim, LoRA params) and encoder_config_class. The .pth is the state_dict for the corresponding ImageEncoderWithHead wrapper.

Quick-start

import anndata as ad
import torch
from cp_bg_bench_model.models._export import load_image_encoder_with_head

a = ad.read_h5ad("01_JUMP_DINO_ECFP4_C/aggregated.h5ad")
emb = a.obsm["X_pca_harmony"]  # (n_wells, 50)

enc = load_image_encoder_with_head("01_JUMP_DINO_ECFP4_C/exports/imgenc_finetuned.json").eval()
with torch.no_grad():
    z = enc(torch.randn(1, 5, 224, 224))  # (1, 128)

Source datasets and licensing

Dataset	Upstream license
JUMP-CP	CC BY 4.0
RxRx1	CC BY-NC-SA 4.0
RxRx3-core	CC BY-NC-SA 4.0

Embeddings are derivative works. The most restrictive upstream license (CC BY-NC-SA 4.0) applies to this repo: non-commercial, share-alike, attribution.

Intended use

Research on representation learning for biological image data and cross-batch generalization. Not validated for clinical, diagnostic, or biosecurity applications.

Anonymous review

This release is anonymized for double-blind review. Please direct correspondence through the OpenReview discussion thread.

Citation

@inproceedings{anon_cpbgbench_2026,
  title={cp-bg-bench: a cross-batch generalization benchmark for cell painting representations},
  author={Anonymous},
  booktitle={NeurIPS Datasets and Benchmarks},
  year={2026},
  note={Under review.}
}

Downloads last month: 49

Total file size:

50.1 GB