Alyon-AI/UMA-Dataset_VLM-Engine-v1

UMA-IA/PYXIS-Engine-v1

Authors

• Youri LALAIN, Engineering student at French Engineering School ECE
• Lilian RAGE, Engineering student at French Engineering School ECE

Dataset Summary

The UMA-IA/PYXIS-Engine-v1 is a specialized dataset designed for training vision-language models in the field of aerospace and aeronautical engineering. It consists of high-quality images of aircraft engine components paired with detailed captions identifying and describing the visible parts. This dataset enables models to learn to recognize and analyze various engine components, making it ideal for fine-tuning vision-language models for technical visual recognition and analysis tasks in the aerospace industry.

Dataset Details

• Splits:
  • Train: Complete dataset for model training
• Columns:
  • image: The image file of an aircraft engine component or cross-section
  • caption: Detailed description of visible components in the image
  • image_id: Unique identifier for each image
  • cui: Technical classification identifier

Dataset Structure

The dataset's primary focus is on providing high-quality annotated images of aircraft engine components with detailed technical descriptions. Each entry contains:

1. An image showing aerospace engine components from various angles and cross-sections
2. Detailed captions identifying components such as:
   • Soufflante (Fan)
   • Aubes (Blades)
   • Rotor
   • Stator
   • Compresseur (Compressor)
   • And other critical engine components

Example Entries

image	caption	image_id
[Engine Image]	Composants visibles: - Soufflante - Aubes de soufflante - Rotor de soufflante...	001269777_896x598_c_mirror
[Engine Image]	Composants visibles: - Soufflante - Aubes - Rotor - Stator - Compresseur...	001269777_896x598_c_original
[Engine Image]	Composants visibles: - Soufflante - Aubes - Rotor - Stator - Compresseur...	001269777_896x598_c_segment1

Applications

This dataset is particularly valuable for:

• Training vision models to recognize aerospace engine components
• Developing diagnostic tools for engine maintenance
• Creating educational resources for aerospace engineering
• Enhancing technical documentation with automatic component recognition
• Supporting quality control processes in engine manufacturing

How to Use

You can load this dataset using the Hugging Face datasets library:

from datasets import load_dataset
dataset = load_dataset("UMA-IA/PYXIS-Engine-v1")

# Access the first sample
print(dataset["train"][0]["caption"])

# Display an image (if in a notebook environment)
from PIL import Image
import matplotlib.pyplot as plt

img = dataset["train"][0]["image"]
plt.figure(figsize=(10, 8))
plt.imshow(img)
plt.axis('off')
plt.title(dataset["train"][0]["caption"])
plt.show()