PixMo-Clocks

PixMo-Clocks is a collection of virtual watch faces and time annotations. The repo supplies the meta-data used to build the data but does not directly contain the images, which are from facer.io.

PixMo-Clocks is a part of the PixMo dataset collection and was used to train the Molmo family of models

Quick links:

• 📃 Paper
• 🎥 Blog with Videos

Loading

data = datasets.load_dataset("allenai/pixmo-clocks", split="train")

Data Format

The data includes:

• watchface_id: The watchface id to use to generate the image
• watchface_name: The name of the watchface used
• watch_time: The time the watch was set to, not all the details of the time will be visible on the watchface, so the the following two fields are needed.
• time_format: The format of the watch, can be:
  • "No time visible": The time cannot be read at all, we still include these examples as no-answer training examples
  • "Without AM/PM": AM/PM cannot be determined
  • "With AM/PM": AM/PM can be determined (either because the watch shows military time, or shows a AM/PM indicator somehow)
• shows_seconds: Whether seconds are shown

For convenience, this dataset includes a few fields derived from this data about what is visible on the watch:

• hour_visible: The hour visible on the watch, between 0 and 23, -1 means not visible, 0 is 12:00am. If AM/PM cannot be determined the hour will be between 0 and 11
• minute_visible: The minute the watch should be set to, between 0 and 59, -1 means not visible
• second_visible: The second the watch should be set to, between -1 and 59, -1 means not visible
• text: A text string of the time visible on the watch

Downloading Images

Getting images requires downloading watchface images set to different times from facer.io. We will include information about how to generate the watch images as part of our code release.

Splits

The data is divided into validation and train splits. These splits are unofficial because we do not use this data for evaluation anyway. However, they reflect what was used when training the Molmo models, which were only trained on the train split.

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.