Category ➡️ Data Science
Subcategory ➡️ Computer Vision Engineer
Difficulty ➡️ Easy
Objetivo del reto ➡️ Este desafío ha sido creado para poner en práctica los conocimientos adquiridos en nuestro curso: Deep Learning A-Z.
In the ever-evolving field of computer vision, object detection stands as a cornerstone, enabling countless applications from autonomous vehicles to surveillance systems. VisionTech calls upon AI researchers and developers to engage in the Object Detection Challenge. The mission is to apply cutting-edge computer vision techniques to identify and localize objects within digital images. Leveraging the comprehensive COCO Dataset, this challenge aims to push the boundaries of what's possible in object detection, providing invaluable insights into both academic research and practical applications.
Participants will utilize the "COCO Dataset," renowned for its diversity, size, and annotation quality, including:
Images across a variety of contexts
Annotations for object detection (bounding boxes, categories)
This dataset is crucial for developing models capable of understanding and interpreting complex visual scenes.
For the training dataset:
Implement suitable data preprocessing and augmentation techniques to enhance model performance and generalization capabilities.
Develop and train a computer vision model capable of object detection. While you're free to select the architecture, popular choices include CNN-based models like Faster R-CNN, YOLO (You Only Look Once), or SSD (Single Shot Multibox Detector).
The repository structure is provided and must be adhered to strictly:
|__README.md
|__requirements.txt
|
|__data
| |__train
| | |__labels.json
| |__test
| |__mapping.csv
|
|__src
| |__data_processing.py
| |__model_training.py
| |__model_prediction.py
| |__utils.py
|
|__models
| |__model.pkl
|
|__predictions
|__example_predictions.json
|__predictions.json
The predictions
folder should contain the predictions.json
file with your model's predicted insect categories. The mapping.csv
contains the relation between each category and their category id.
Task 1: Create a model using computer vision techniques that effectively identifies and localizes objects in digital images. The solution should demonstrate a keen understanding of object detection fundamentals and innovative approaches to tackling the challenge.
Submit a predictions.json file containing the objects detected by your model. Ensure the file is formatted correctly, with image id as keys and as values a list of the bounding boxes with the category of the object detected. For each image there may be one or more objects to detect. predictions.json:
{
"target": {
"289343": [
{"bbox": [188.1,200.23,90.97,279.77],"category": 11},
{"bbox": [22.1,515.6,231.4,66.8],"category": 42},
...
],
"61471": [
{"bbox": [474.07,395.93,38.65,77.67],"category": 24},
...
],
...
}
}
The performance of your object detection model will be evaluated based the mAP (mean Average Precision) metric. This metric provide a comprehensive assessment of the model's accuracy and its ability to detect objects across various conditions.
⚠️ Please note:
All submissions might undergo a manual code review process to ensure that the work has been conducted honestly and adheres to the highest standards of academic integrity. Any form of dishonesty or misconduct will be addressed seriously, and may lead to disqualification from the challenge.
Ensure that all data manipulation and model training strictly utilize the libraries mentioned in requirements.txt.
Q1: What is the goal of the Object Detection Challenge?
A1: To develop a robust computer vision system that accurately identifies and localizes objects within digital images.
Q2: What type of data will I be working with?
A2: The challenge uses the COCO Dataset, which includes images with annotated objects for detection.
Q3: Are there any recommended models or algorithms?
A3: It is recommended to explore CNN-based architectures like Faster R-CNN, YOLO, or SSD, ensuring you adhere to the libraries from requirements.txt.
Q4: How will the system's performance be evaluated?
A4: System performance will be evaluated using mAP, focusing on the accuracy and reliability of object detections.
Timeline
01
Start the challenge & clone the repository
02
Solve the challenge & submit your solution
Next action: