| --- |
| license: cc-by-nc-sa-4.0 |
| extra_gated_fields: |
| full_name: |
| type: text |
| label: What is your full name? |
| required: true |
| email: |
| type: text |
| label: What is your email address? |
| required: true |
| company: |
| type: text |
| label: Which company or institution are you affiliated with? |
| required: false |
| intended_use: |
| type: text |
| label: Please describe your intended use of this model. |
| required: true |
| agreement: |
| type: text |
| label: >- |
| Type "I agree" to confirm you have read and accept the license and usage |
| conditions. |
| required: true |
| tags: |
| - objectdetection |
| - crowdhuman |
| - person |
| - head |
| - ai |
| --- |
| |
| --- |
|
|
|  |
|
|
|
|
| # Model Card for `rebotnix/rb_crowhuman` |
| |
| > **Person and Head Detection** – Trained by KINEVA, Built by REBOTNIX, Germany |
| Current State: in production and re-training. |
| |
| --- |
| |
| `rb_crowhuman` is a specialized object detection model trained on the **CrowdHuman** dataset to detect **persons** and **heads** in densely crowded scenes. Designed for reliable detection in challenging conditions with heavy occlusion, overlapping bodies, and varying crowd densities, this model is well suited for crowd analysis, people counting, and safety monitoring in public spaces. |
|
|
| Developed and maintained by **REBOTNIX**, Germany, https://rebotnix.com |
|
|
| # About KINEVA |
| KINEVA is an automated training platform based on the MCP Agent system. It regularly delivers new visual computing models, all developed entirely from scratch. This approach enables the creation of customized models tailored to specific client requirements, which can be retrained and re-released as needed. The platform is particularly suited for applications that demand flexibility, adaptability, and technological precision—such as industrial image processing, smart city analytics, or automated object detection. |
|
|
| KINEVA is continuously evolving to meet the growing demands in the fields of artificial intelligence and machine vision. https://rebotnix.com/en/kineva |
|
|
| --- |
|
|
| ## Example Predictions |
|
|
| | Input Image | Detection Result | |
| |-------------|------------------| |
| | <img src="./example_crowdhuman1.jpg" width="300"/> | <a href="./rb_crowhuman/blob/main/output_1.jpg" target="_blank"> <img src="./output_1.jpg" width="300"/></a> | |
| | <img src="./example_crowdhuman2.jpg" width="300"/> | <a href="./rb_crowhuman/blob/main/output_2.jpg" target="_blank"> <img src="./output_2.jpg" width="300"/></a> | |
|
|
| --- |
|
|
| ## Model Details |
|
|
| - **Architecture**: KINEVA SILVER *(custom training head with optimized anchor boxes)* |
| - **Task**: Person and Head Detection (2 classes: person, head) |
| - **Trained on**: CrowdHuman dataset |
| - **Format**: PyTorch `.pth` + ONNX and TRT export available on request |
| - **Training Framework**: PyTorch + KINEVA + custom augmentation |
|
|
| --- |
|
|
| We're happy to **license or provide access to all intermediate weights** for research or further development purposes. Please feel free to reach out. |
|
|
| ## Dataset |
|
|
| The model was trained on the **CrowdHuman** dataset, featuring: |
|
|
| - Over 15,000 images with highly crowded scenes |
| - Annotations for full body (person) and head bounding boxes |
| - Heavy occlusion and overlapping individuals |
| - Diverse indoor and outdoor environments |
|
|
| More on CrowdHuman: [https://www.crowdhuman.org](https://www.crowdhuman.org) |
|
|
| --- |
|
|
| ## Intended Use |
|
|
| | Intended Use | Not Intended Use | |
| |----------------|---------------------| |
| | Person and head detection in crowded scenes | Surveillance without human review | |
| | Crowd density estimation and people counting | Military / lethal applications | |
| | Public safety and occupancy monitoring | Real-time safety-critical decisions without human oversight | |
| | Retail analytics and footfall analysis | Individual identification or tracking | |
|
|
| --- |
|
|
| ## Limitations |
|
|
| - May yield false positives in scenes with mannequins or posters depicting people |
| - Not fine-tuned for thermal or night vision imagery |
| - Heavy occlusion in extremely dense crowds may reduce detection accuracy |
| - Performance may degrade on low-resolution or distant subjects |
|
|
| --- |
|
|
| ## Usage Example |
|
|
| ```python |
| |
| from kineva import KINEVA |
| |
| #initialize model |
| model = KINEVA(model="models/kineva_crowdhuman.pth") |
| |
| #run inference on image |
| final_boxes, final_scores, final_labels = model.detect("example_crowdhuman1.jpg", threshold=0.35) |
| |
| #draw detection |
| model.draw(final_boxes, final_scores, final_labels, output_path="./output_1.jpg") |
| |
| ``` |
|
|
| --- |
|
|
| ## Contact |
|
|
| For commercial use or re-training this model support, or dataset access, contact: |
|
|
| **REBOTNIX** |
| Email: [communicate@rebotnix.com](mailto:communicate@rebotnix.com) |
| Website: [https://rebotnix.com](https://rebotnix.com) |
|
|
| --- |
|
|
| ## License |
|
|
| This model is released under **CC-BY-NC-SA** unless otherwise noted. For commercial licensing, please reach out to the contact email. |
|
|
| --- |
|
|
