README.md

---
license: mit
task_categories:
- image-classification
- visual-question-answering
- image-to-text
tags:
- 3d-printing
- manufacturing
- quality-control
- vision-language
size_categories:
- 1K<n<10K
pretty_name: TL-Caxton - 3D Printing Quality Assessment Dataset
---

# 3D Printing Nozzle Images Dataset

### Dataset Summary

- **Task**: Vision-based flow rate estimation and extrusion quality assessment
- **Domain**: Additive Manufacturing / 3D Printing
- **Data Type**: RGB images with numerical annotations
- **Total Samples**: 4,048 images
  - Training: 3,407 samples
  - Validation: 331 samples
  - Test: 310 samples

### Supported Tasks

1. **Flow Rate Regression**: Predict the flow rate percentage from camera images of the printing process
2. **Extrusion Quality Classification**: Classify prints as under-extruded (<90%), good extrusion (90-110%), or over-extruded (>110%)
3. **Vision-Language Modeling**: Generate natural language descriptions of print quality from images
4. **Visual Question Answering**: Answer questions about print parameters and quality from images

## Dataset Structure

### Data Fields

Each sample contains:

- **`img_path`** (string): Filename of the camera image
- **`flow_rate`** (float): Flow rate percentage value (ranging from ~39% to ~265%)
- **`nozzle_tip_x`** (int): X-coordinate of nozzle tip position in pixels
- **`nozzle_tip_y`** (int): Y-coordinate of nozzle tip position in pixels

### Data Splits

| Split | Samples | Percentage |
|-------|---------|------------|
| Train | 3,407   | 84.2%      |
| Validation | 331 | 8.2%       |
| Test  | 310     | 7.6%       |
| **Total** | **4,048** | **100%** |

### Qualitative Descriptions

The dataset includes JSON template files for generating natural language descriptions:

- **`general_statements.json`**: General observations about the 3D printing nozzle and process
- **`qual_good_extrusion.json`**: Descriptions of good extrusion quality (flow rate 90-110%)
- **`qual_under_extrusion.json`**: Descriptions of under-extrusion issues (flow rate < 90%)
- **`qual_over_extrusion.json`**: Descriptions of over-extrusion issues (flow rate > 110%)
- **`quant_templates.json`**: Templates for stating quantitative flow rate values

These templates enable synthetic generation of diverse natural language annotations for vision-language training.

## Usage

### Loading the Dataset

```python
from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("cemag/tl-caxton")

# Access individual splits
train_data = dataset['train']
val_data = dataset['validation']
test_data = dataset['test']

# Example: Access a sample
sample = train_data[0]
print(f"Flow rate: {sample['flow_rate']}%")
print(f"Nozzle position: ({sample['nozzle_tip_x']}, {sample['nozzle_tip_y']})")
```

### Using with PyTorch

```python
from torch.utils.data import DataLoader
from PIL import Image
import os

class CIPHERDataset:
    def __init__(self, dataset, image_dir, transform=None):
        self.dataset = dataset
        self.image_dir = image_dir
        self.transform = transform
    
    def __len__(self):
        return len(self.dataset)
    
    def __getitem__(self, idx):
        sample = self.dataset[idx]
        img_path = os.path.join(self.image_dir, sample['img_path'])
        image = Image.open(img_path).convert('RGB')
        
        if self.transform:
            image = self.transform(image)
        
        return {
            'image': image,
            'flow_rate': sample['flow_rate'],
            'nozzle_tip': (sample['nozzle_tip_x'], sample['nozzle_tip_y'])
        }

# Create dataset and dataloader
train_dataset = CIPHERDataset(train_data, 'images/', transform=your_transform)
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
```

### Vision-Language Training

```python
from data_utils import synthesize_answer, format_data

# Generate a natural language description for a sample
sample = train_data[0]
description = synthesize_answer(sample, general=True, quant=True, qual=True)
print(description)

# Example output:
# "This is the nozzle of a 3D printer. The observed flow rate is approximately 
#  100%. Good extrusion occurs when a 3D printer delivers the exact amount of 
#  filament needed, resulting in strong, accurate, and visually appealing prints."
```

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{tl_caxton,
  title={tl-Caxton: 3D Printing Quality Assessment Dataset},
  author={cemag},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/cemag/tl-caxton}}
}
```

```bibtex
@article{MargadjiPattinson2025HybridReasoning,
  title   = {Hybrid Reasoning for Perception, Explanation, and Autonomous Action in Manufacturing},
  author  = {Margadji, Christos and Pattinson, Sebastian W.},
  year    = {2025},
  note    = {arXiv:2506.08462},
  url     = {https://arxiv.org/abs/2506.08462}
}
```

## License

This dataset is released under the MIT License.

## Contact

For questions or issues regarding this dataset, please open an issue on the dataset repository or email at cm2161@cam.ac.uk