People Computer Vision Project
Updated 2 months ago
Metrics
Use this model to detect people across diverse scenarios. It is an excellent way to start detecting people from scratch or a great starting checkpoint for a more targeted use case!
For this model, we curated a dataset from COCO-2017 and various projects from Roboflow Universe to create a model that generalizes well. To start, we sampled 25k images from COCO-2017 and included annotations that are improved upon the original in a few different areas: Improved annotation in crowds of people. Clusters of objects are relabeled as their own unique items. Significant reduction in number of very small objects - 10x10 pixels or less
Additionally we selected data that ranged in type, context, annotation size, annotation count and cameras from Roboflow Universe:
- person_camera_security1 (CC BY 4.0 license)
- Human Action Recognition 2000 (CC BY 4.0 license)
- People Detection (CC BY 4.0 license)
- contador-de-gente teste 3 (CC BY 4.0 license)
- OD3 (CC BY 4.0 license)
- MOT17-03-DPM
- The Curve (CC BY 4.0 license)
- pedestrain safety (CC BY 4.0 license)
After the data was collected we used the following preprocessing pipeline for feeding data to the model:
- Resize to 640x640 with stretch to standardize all image sizes for input to the model.
- Keep 10% of null samples to prevent overfitting to certain backgrounds and reduce false positives.
Use This Trained Model
Try it in your browser, or deploy via our Hosted Inference API and other deployment methods.
Build Computer Vision Applications Faster with Supervision
Visualize and process your model results with our reusable computer vision tools.
Cite This Project
If you use this dataset in a research paper, please cite it using the following BibTeX:
@misc{
people-kahye_dataset,
title = { People Dataset },
type = { Open Source Dataset },
author = { Lennys Workspace },
howpublished = { \url{ https://universe.roboflow.com/lennys-workspace-gabu9/people-kahye } },
url = { https://universe.roboflow.com/lennys-workspace-gabu9/people-kahye },
journal = { Roboflow Universe },
publisher = { Roboflow },
year = { 2024 },
month = { oct },
note = { visited on 2024-12-21 },
}