---
library_name: transformers
tags:
- object-detection
- license-plate
license: apache-2.0
datasets:
- ariG23498/license-detection-paligemma
base_model:
- google/gemma-3-4b-pt
pipeline_tag: object-detection
---

# Gemma 3 4B Fine-Tuned for Object Detection

This model is a fine-tuned version of Gemma 3 4B for license plate object detection.


| Detected License Plates (Sample 1) | Detected License Plates (Sample 2) |
| :--------------------------------: | :--------------------------------: |
|      ![image/png](https://cdn-uploads.huggingface.co/production/uploads/61929226ded356549e20c5da/1BgYT_F9V22ULMJ4yYEdn.png)     |      ![image/png](https://cdn-uploads.huggingface.co/production/uploads/61929226ded356549e20c5da/jpZjRKEfOHu5qqiYKVXp-.png)     |
|      ![image/png](https://cdn-uploads.huggingface.co/production/uploads/61929226ded356549e20c5da/-SrbuYyr0HvfY8vsIr_-i.png)    |    ![image/png](https://cdn-uploads.huggingface.co/production/uploads/61929226ded356549e20c5da/6KTp4BanHbbf3NopRn4eX.png)|

## Model Details

### Model Description

This model aims to prove that VLMs **NOT previously** trained for object detection and **without previous knowledge** of location tokens (`<locXXXX>`) can still be fine tuned for object detection out of the box. This is an experimental model.

- **Developed by:** [Aritra Roy Gosthipaty](https://huggingface.co/ariG23498) and [Sergio Paniego](https://huggingface.co/sergiopaniego)
- **Finetuned from model:** [gemma-3-4b-pt](https://huggingface.co/google/gemma-3-4b-pt)

### Model Sources

- [**Repository:**](https://github.com/ariG23498/gemma3-object-detection)
- [**HF Space:**](https://huggingface.co/spaces/ariG23498/gemma3-license-plate-detection)
- [**Collection:**](https://huggingface.co/collections/ariG23498/gemma-3-object-detection-682469cb72084d8ab22460b3)
- [**Dataset:**](https://huggingface.co/datasets/ariG23498/license-detection-paligemma)

## Uses

Follow these steps to configure, train, and run predictions (using the code repository):

1. Configuration (`config.py`): All major parameters are centralized here. Before running any script, review and adjust these settings as needed.
2. Training (`train.py`): This script handles the fine-tuning process.
3. Running inference (`infer.py`): Run this to visualize object detection.

## Citation

If you use our work, please cite us:

```
@misc{gosthipaty_gemma3_object_detection_2025,
  author = {Aritra Roy Gosthipaty and Sergio Paniego},
  title = {Fine-tuning Gemma 3 for Object Detection},
  year = {2025},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/ariG23498/gemma3-object-detection.git}}
}
```