Datasets:

CorentinDumery
/

MixCount

license: cc-by-4.0
dataset_info:
  features:
    - name: image
      dtype: image
    - name: file_name
      dtype: string
    - name: total_count
      dtype: int64
    - name: num_classes
      dtype: int64
    - name: class_names
      sequence: string
    - name: class_counts
      sequence: int64
    - name: class_descriptions
      sequence: string
    - name: objects
      struct:
        - name: bbox
          sequence:
            sequence: float64
        - name: category
          sequence: int64
        - name: global_category_id
          sequence: int64
configs:
  - config_name: default
    data_files:
      - split: train
        path: parquet_data/train-*
      - split: validation
        path: parquet_data/val-*
      - split: test
        path: parquet_data/test-*
tags:
  - counting
  - synthetic
  - computer-vision
  - open-vocabulary
  - segmentation
pretty_name: MixCount
task_categories:
  - image-classification
  - object-detection

The MixCount Dataset: Bridging the Data Gap for Open-Vocabulary Object Counting

Corentin Dumery* · Niki Amini-Naieni* · Shervin Naini · Pascal Fua
_{EPFL · University of Oxford · Northwestern University · (* equal contribution)}

MixCount sample scenes (2×6 grid)

MixCount is a large-scale synthetic dataset for mixed-object, open-vocabulary counting, the setting that dominates industrial inspection and sorting, but breaks current counting models. Our automatic generation pipeline produces pixel-perfect labels, text prompts at several levels of detail, and visual exemplars at scale.

58Kcounting scenes

1,522object classes

4M+counting instances

−18.3%MAE on PairTally (train)

−20.14%MAE on FSC-147 (train)

Usage

import matplotlib.pyplot as plt
import matplotlib.patches as patches
from datasets import load_dataset

# Load the streaming dataset
dataset = load_dataset("CorentinDumery/MixCount", split="train", streaming=True)
example = next(iter(dataset))

# Print image-level metadata
print(f"Total Count: {example['total_count']}")
print(f"Number of Classes: {example['num_classes']}\n")

for name, count, desc in zip(example['class_names'], example['class_counts'], example['class_descriptions']):
    print(f" - {name}: {count} instance(s)")
    print(f"   Description: {desc}")

# Extract object and class data
objects = example['objects']
class_names = example['class_names']

# Set up the plot
fig, ax = plt.subplots(1, figsize=(10, 8))
ax.imshow(example['image'])

# Extract lists, falling back to None if global_category_id isn't in the parquet yet
bboxes = objects['bbox']
local_categories = objects['category']
global_categories = objects.get('global_category_id', [None] * len(bboxes))

# Draw bounding boxes and labels
for bbox, local_idx, global_id in zip(bboxes, local_categories, global_categories):
    x, y, w, h = bbox
    
    # Color based on the local index so instances of the same class share a color
    color = plt.colormaps['tab20'](local_idx % 20)
    
    # Format the label string
    if global_id is not None:
        label_name = f"{class_names[local_idx]} (Global ID: {global_id})"
    else:
        label_name = f"{class_names[local_idx]}"
    
    # Draw Bounding Box
    rect = patches.Rectangle(
        (x, y), w, h,
        linewidth=2,
        edgecolor=color,
        facecolor='none',
    )
    ax.add_patch(rect)
    
    # Draw Text Label above the bounding box
    ax.text(
        x, y - 5, 
        label_name, 
        color='white', 
        fontsize=8,
        bbox=dict(facecolor=color, edgecolor='none', alpha=0.8)
    )

plt.axis('off')
plt.tight_layout()
plt.show()

Overview

Object counting models often struggle in mixed-object scenes. Common failure modes include:

(a) Distinguishing visually similar objects (e.g. big marbles in PairTally)
(b) Recognizing self-similar components as a single entity (e.g. counting pairs of sunglasses rather than lenses)
(c) Ignoring repetitive background patterns and focusing on the queried object class

MixCount combines the scale of synthetic datasets with the photorealism of real-world 3D captures while targeting these failure modes. Training on MixCount yields about 20% lower error on recent open-vocabulary counting benchmarks.

Training on MixCount improves CountGD++ on PairTally, FSC-147, and MixCount

Dataset overview

	FSC-147	PairTally	MCAC	MixCount
Multiple object types per image		✓	✓	✓
Fine-grained text prompts		✓		✓
External visual exemplars				✓
Segmentation & bounding boxes			✓	✓
# images	6,135	681	20K	58,000
# object classes	147	98	343	1,522

Visual & text inputs. Multiple visual exemplars per object (external crops and in-scene crops at different scales), together with short, concise, and detailed text descriptions for flexible open-vocabulary counting prompts.

MixCount exemplars and tiered text descriptions

Dense annotations. Pixel-perfect counting supervision plus instance and class segmentations, bounding boxes, depth, and normal maps.

MixCount dense annotations

Automatic generator. Objects, distractors, environment, and camera placement are sampled procedurally to create photorealistic training scenes from high-quality real-world captures of objects, materials, and lighting.

MixCount data generation pipeline

See the project page and paper for additional details.

Citation

@article{dumery2026mixcount,
  title   = {The MixCount Dataset: Bridging the Data Gap for Open-Vocabulary Object Counting},
  author  = {Dumery, Corentin and Amini-Naieni, Niki and Naini, Shervin and Fua, Pascal},
  journal = {arXiv preprint arXiv:2605.18063},
  year    = {2026}
}

Acknowledgements

We thank DTC, VasTextures, LavalIndoor, and PolyHaven, as well as the Blender Foundation. We also thank Andrew Zisserman for insightful discussions. This work is partially funded by the Swiss National Science Foundation, an AWS Studentship, the Reuben Foundation, a Qualcomm Innovation Fellowship (mentors: Dr Farhad Zanjani and Dr Davide Abati), and the AIMS CDT program at the University of Oxford.