license: odc-by
dataset_info:
features:
- name: image_url
dtype: string
- name: image_sha256
dtype: string
- name: points
list:
- name: x
dtype: float64
- name: 'y'
dtype: float64
- name: count
dtype: int64
- name: label
dtype: string
- name: collection_method
dtype: string
splits:
- name: train
num_bytes: 668565775
num_examples: 2376222
download_size: 198336771
dataset_size: 668565775
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
PixMo-Points
PixMo-Points is a dataset of images paired with referring expressions and points marking the locations the referring expression refers to in the image. It was collected using human annotators and contains a diverse range of points and expressions, with many high-frequency (10+) expressions.
PixMo-Points is a part of the PixMo dataset collection and was used to provide the pointing capabilities of the Molmo family of models
Quick links:
Loading
data = datasets.load_dataset("allenai/pixmo-points", split="train")
Data Format
Images are stored as URLs that will need to be downloaded separately. Note URLs can be repeated in the data.
The points field contains the x, y coordinates specified in pixels.
The label field contains the string name of what is being pointed at, this can be a simple object name or a more complex referring expression.
The collection_method field specifies whether the image was chosen to target high-frequency counting ("counting") or general pointing ("pointing").
Image Checking
Image hashes are included to support double-checking that the downloaded image matches the annotated image. It can be checked like this:
from hashlib import sha256
import requests
example = data[0]
image_bytes = requests.get(example["image_url"]).content
byte_hash = sha256(image_bytes).hexdigest()
assert byte_hash == example["image_sha256"]
License
This dataset is licensed under ODC-BY-1.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines.