The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
AITZ-Aug: Enhanced CoaT Dataset with Data Augmentation by GPT-4o
AITZ-Aug is an enhanced version of the CoaT dataset, augmented using GPT-4o. The dataset is provided in a 7Z archive containing the following files and directories:
android_in_the_zoo: Data after augmentation.android_in_the_zoo.zip: Compressed file containing the augmented dataset.android_in_the_zoo_backup: Original dataset without augmentation.noisy: Contains the AITZ-noisy data from Mobile-Bench-v2.
Augmentation Details
For each page in the original dataset, AITZ-Aug constructs 15 additional QA pairs divided into the following three types:
Functionality Questions and Answers:
Questions about the function of specific elements in the image.Grounding Questions and Answers:
Questions about the coordinates and approximate location of targets within the image.Partial Detailed Information Questions and Answers:
Questions that focus on the structural relationships, types, styles, and other detailed information of multiple elements.
Purpose of AITZ-Aug
These QA pairs serve as a means to achieve fine-grained page understanding. They aim to:
- Prevent model overfitting.
- Enhance compatibility with DPO (Direct Preference Optimization) sampling during training.
This structured augmentation provides richer annotations, making the dataset more suitable for training advanced models for fine-grained understanding tasks.
- Downloads last month
- 3