ABC Easy as 123: A Blind Counter for Exemplar-Free
Multi-Class Class-agnostic Counting


MCAC: A Multi-Class Class-Agnostic Counting Dataset

Michael A. Hobley, Victor A. Prisacariu
University of Oxford

Paper Code Dataset


Class-agnostic counting methods enumerate objects of an arbitrary class, providing tremendous utility in many fields. Prior works have limited usefulness as they require either a set of examples of the type to be counted or that the image contains only a single type of object. A significant factor in these shortcomings is the lack of a dataset to properly address counting in settings with more than one kind of object present. To address these issues, we propose the first Multi-class, Class-Agnostic Counting dataset (MCAC) and A Blind Counter (ABC123), a method that can count multiple types of objects simultaneously without using examples of type during training or inference. ABC123 introduces a new paradigm where instead of requiring exemplars to guide the enumeration, examples are found after the counting stage to help a user understand the generated outputs. We show that ABC123 outperforms contemporary methods on MCAC without the requirement of human in-the-loop annotations. We also show that this performance transfers to FSC-147, the standard class-agnostic counting dataset.

ABC123 enumerates instances of multiple types of objects simultaneously without needing exemplar images. Not only does ABC123 not require exemplar images, it finds examples of type to aid a user in understanding the types it has counted.


MCAC Example. Each object in the RGB image has an associated: Model ID, Class ID, Center Coordinate, Bounding Box and Occlusion Percentage.

More examples can be seen here.

This dataset is a set of synthetic images of objects for the purpose of multi-class few-shot or class-agnostic counting.

The classes of objects present in the Train, Test and Val splits are mutually exclusive, and where possible aligned with the class splits in FSC-133.

Models are taken from ShapeNetSem. The original model IDs and manually verified category labels are preserved.

Benchmark Results

Val Test
Mean N/A 39.87 53.56 3.07 11.40 42.67 59.68 2.79 10.93
Median N/A 36.25 58.15 1.51 6.70 39.81 65.36 1.38 6.73
FamNet 3 24.76 41.12 1.12 6.86 26.40 45.52 1.04 6.87
BMNet 3 15.83 27.07 0.71 4.97 17.29 29.83 0.75 6.08
CounTR 3 15.07 26.26 0.63 4.79 16.12 29.28 0.67 5.71
ABC123 0 8.96 15.93 0.29 2.02 9.52 17.64 0.28 2.23

File hierarchy

      ├── metadata
      │   ├── generate_shapenetsem_split_new.ipynb
      │   └── shapenetsem001_train_test_new.json
      ├── test
      ├── train
      │   ├── 1511489148409439
      │   ├── 3527550462177290
      │   |   ├──img.png
      │   |   ├──info.json
      │   |   ├──seg.png
      │   |   └──seginds
      │   |      ├──0.png
      │   |      ├──1.png
      │   |      └── ...
      │   ├──4109417696451021
      │   └── ...
      └── val


        title={ABC Easy as 123: A Blind Counter for Exemplar-Free Multi-Class Class-agnostic Counting},
        author={Hobley, Michael and Prisacariu, Victor},
        journal={arXiv preprint arXiv:2309.04820},