Datasets:

marloweee
/

BARE_grefumd_test_partitioning

license: apache-2.0
task_categories:
  - object-detection
language:
  - en
tags:
  - referring-expression-comprehension
  - visual-grounding
  - refcocog
  - fine-grained-evaluation
size_categories:
  - 1K<n<10K

RefCOCOg-UMD Test Partitioning

This dataset provides fine-grained evaluation splits for the RefCOCOg-UMD test set, designed for Referring Expression Comprehension (REC) tasks. Each sample is partitioned by difficulty level and referential type to enable more detailed performance analysis.

Dataset Description

Total Samples: 9,482 (by difficulty) / 19,334 (by referential type, with overlaps)
Source: RefCOCOg-UMD test set
Format: PyTorch .pth files

Data Structure

Each .pth file contains a list of tuples with the following structure:

import torch

data = torch.load("Easy.pth")
for item in data:
    img_name, hw_dict, bbox, phrase, obj_mask = item
    # img_name: str - Image filename
    # hw_dict: dict - Height and width info {"height": int, "width": int}
    # bbox: list - Bounding box [x1, y1, x2, y2]
    # phrase: str - Referring expression text
    # obj_mask: tensor/array - Object segmentation mask

File Organization

By Difficulty Level

Easy.pth - Simple referring expressions (4,109 samples)
Medium.pth - Moderate complexity (4,517 samples)
Hard.pth - Complex expressions (856 samples)

By Referential Type × Difficulty

{Type}_Easy.pth, {Type}_Medium.pth, {Type}_Hard.pth
Types: Attribute, Relation, Logic, Ambiguity, Perspective

Statistics for Each File

Filename	Avg. Length	Samples
Attribute_Easy.pth	35.92	3,076
Attribute_Medium.pth	48.74	3,431
Attribute_Hard.pth	50.90	563
Relation_Easy.pth	37.86	2,673
Relation_Medium.pth	47.51	4,092
Relation_Hard.pth	51.95	709
Logic_Easy.pth	49.92	172
Logic_Medium.pth	53.27	376
Logic_Hard.pth	61.68	118
Ambiguity_Easy.pth	32.35	174
Ambiguity_Medium.pth	42.62	1,659
Ambiguity_Hard.pth	45.64	783
Perspective_Easy.pth	30.07	299
Perspective_Medium.pth	44.29	955
Perspective_Hard.pth	52.14	254
Easy.pth	33.85	4,109
Medium.pth	46.19	4,517
Hard.pth	47.87	856

Statistics for Difficulty Levels and Referential Types

STATISTICS OF THE FINE-GRAINED EVALUATION SPLITS ON REFCOCOG-UMD TEST SET. EACH SAMPLE IS ASSIGNED ONE DIFFICULTY LEVEL, WHILE REFERENTIAL CATEGORIES MAY OVERLAP ACROSS SAMPLES.

Category	Subset	Avg. Length	Samples	Total
Difficulty	Easy	33.85	4,109	9,482
	Medium	46.19	4,517
	Hard	47.87	856
Referential Type	Attribute	43.33	7,070	19,334
	Relation	44.48	7,474
	Logic	53.89	666
	Ambiguity	42.84	2,616
	Perspective	42.79	1,508

Usage

from huggingface_hub import hf_hub_download
import torch

# Download a specific file
file_path = hf_hub_download(
    repo_id="marloweee/BARE_grefumd_test_partitioning",
    filename="Easy.pth",
    repo_type="dataset"
)

# Load and iterate
data = torch.load(file_path)
for img_name, hw_dict, bbox, phrase, obj_mask in data:
    print(f"Image: {img_name}, Query: {phrase}")

License

This dataset is released under the Apache 2.0 License.