ninamoss/sleeetview_agentic_ai_dataset

The SleetView Agentic AI Dataset

The SleetView Agentic AI dataset is a collection of synthetic content automatically generated using Agentic AI

Dataset Details

Dataset Description

The images were generated with a collection of models available under the Apache-2.0 or creativeml-openrail-m licenses. To generate this dataset we used our own agentic implementation given the goal of creating a dataset that can be used to research synthetic content detection. As pioneers in the synthetic content detection realm, we think having a varied sampling of synthetic data is important to determine detection efficiency.

This dataset allows evaluation for the following scenarios:

• Varied aspect ratios including multiple resolutions common in digital systems
• Landscapes, portraits with one or more characters, pets and other animals, automobiles and architecture
• Multiple lighting scenarios including ambient light, spot lights, night time/day time

Also included in the dataset are segmentation masks and metadata generated with the DETR panoptic model: https://huggingface.co/facebook/detr-resnet-101-panoptic

• Shared by: Mendit.AI
• License: creativeml-openrail-m

Uses

This dataset can be useful for the following research areas:

• Synthetic content detection
• Evaluation of the quality of Agentic AI generated synthetic datasets
• Image segementation quality based on different composition and lighting scenarios

Out-of-Scope Use

Please refer to the creativeml-openrail-m license for restrictions

Dataset Structure

The dataset contains 248 images with the following structure:

• Image
• Associated segmentation mask
• Segmentation metadata formatted as json

Dataset Creation

Agentic AI Generation

An in depth explanation of our approach to agentic generation of synthetic content can be found here: https://menditai.substack.com/p/the-night-the-dataset-appeared-an We opted for a local setup using Ollama and Falcon3 as the LLM powering the agent. Based on our experience with this process we find:

• Instruction tuned LLMs that do not including reasoning are the best for this task

Annotations [optional]

Segmentation masks and metadata were automatically generated using the DETR panoptic model: https://huggingface.co/facebook/detr-resnet-101-panoptic

Citation [optional]

BibTeX:

@article{DBLP:journals/corr/abs-2005-12872, author = {Nicolas Carion and Francisco Massa and Gabriel Synnaeve and Nicolas Usunier and Alexander Kirillov and Sergey Zagoruyko}, title = {End-to-End Object Detection with Transformers}, journal = {CoRR}, volume = {abs/2005.12872}, year = {2020}, url = {https://arxiv.org/abs/2005.12872}, archivePrefix = {arXiv}, eprint = {2005.12872}, timestamp = {Thu, 28 May 2020 17:38:09 +0200}, biburl = {https://dblp.org/rec/journals/corr/abs-2005-12872.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }

@misc{Falcon3, title = {The Falcon 3 Family of Open Models}, url = {https://huggingface.co/blog/falcon3}, author = {Falcon-LLM Team}, month = {December}, year = {2024} }