🧩  Object Detection: Detect pedestrians & vehicles from input images

🛠 Start Solving

🗃 Explore Dataset

🕵🏼‍♀️ What Is Object Detection Puzzle About?

Radar detectors, along with cameras, help navigate the car when visibility is poor. In tough weather conditions like fog or heavy rains, these sensors help in identifying the surrounding environment. They send pulses of radio waves to locate an object and send back signals about the speed and location of that object.

In this puzzle, with radar data from your car, detect different vehicles around you.

💪🏼 What You’ll Learn

In this puzzle, you will learn

1. Multi-class Object Detection
2. How self-driving car detects objects

Let’s get started! 🚀

📝 The Task

The challenge is to use the image dataset to build an automated algorithm to detect individual pedestrians and other kinds of vehicles through bounding boxes.

In machine learning terms: this is multi-class object detection.

👩🏽‍💻 Explore Dataset

The dataset contains information about the radars points generated from a Carla Simulator and whether a dynamic object ( pedestrian, vehicle, etc. ) was within a distance threshold or not.

The individual radar points of single training sample features contain below metadata -

🗂 Dataset Files

The following files are available in the resources section:

• train.zip:( 3000 samples ) This contains images for the training set.

• train.json:( 3000 samples ) This is the train annotations in MS-COCO format.

• test.zip: ( 1000 samples )This contains images for the testing set.

🔏 Evaluation Criteria

The evaluation metrics for this competition are Average Precision ( Primary Score ) @ IoU=0.50:0.95 area all and maxDets set to 100.

🚀 Getting Started

The starter kit breaks down everything from downloading the dataset, loading the libraries, processing the data, creating, training, and testing the model.

Click here to access the basic starter kit. This will share in-depth instructions to

1. Download the necessary files
2. Setup the AIcrow-CLI environment that will help you make a submission directly via a notebook
3. Downloading dataset & importing libraries
4. Preprocessing the dataset
5. Creating the model
6. Setting the model
7. Training the model
8. Submitting the result
9. Uploading the results

Check out the starter kit here.

Submission Format

1. Create an empty list
2. Generate the predictions of individual images
3. Add a dictionary to the list with the following keys & values:
   1. imageid: Id of the image you predicted, it should be +1 the file id. so forex. ( if the file name is 456.jpg, the imageid will be 457
   2. category_id: Id of the predicted class, below the classes and corresponding class.
      1. bicycle: 0
      2. motorcycle: 1
      3. passenger_car: 2
      4. person : 3
   3. bbox: Predicted bounding boxes in (x, y, w, h) format.
   4. score: Probability of the prediction
4. Save the list as a submission.json file.

Find the complete starter kit over here.

🤫 Hint to get started

You can use various image processing techniques such as canny, edge detection, and contours to find bounding boxes & crop the image as per the coordinates. Then creating embedding for the cropped image to use a Random Forest Classifier to predict the object type.

📚 Resource Circle

Check out the full code of Multi-class Object Detection using Detection2 by one of our community members, konstantin_diachkov.

👯‍♀️ Get Help From Community

Hop over to the AIcrowd Blitz discord server to see ongoing discussions about this puzzle.

🙋‍♀️ Subscription Queries

This is one of the many free Blitz puzzles you can access forever. To access more puzzles from various domains from the Blitz Library and to receive a special new puzzle in your inbox every 2 weeks, you can subscribe back to AIcrowd Blitz here.