just-maiyak / image

Projet_image

VISION - Image

An attempt to the Codalab Vision Challenge

Unmodified intro file

This is the Vision project. Since the end of the 20th century, autonomous vehicles have been debated within the scientific community. One of the issue raised are the behavior of the vehicule which depend of the obstacle. In this challenge, we will study the preliminary stage of the Decision, ie the classification of detected entities
(by the cameras of the vehicle for example). To illustrate this problematic, we propose to study the image source CIFAR-10 which groups entities that can interact with the vehicle environment like animals(cat, horse, dog, ...)
and vehicles (bike, car, truck, ...).

Credits: Vincent Boyer, Warren Pons, Ludovic Kun, Qixiang PENG

Prerequisites: Install Anaconda Python 2.7, including jupyter-notebook

Usage:

(1) If you are a challenge participant:

• The file README.ipynb contains step-by-step instructions on how to create a sample submission for the Vision challenge. At the prompt type: jupyter-notebook README.ipynb

• Download the public_data and replace the sample_data with it.

• Modify sample_code_submission to provide a better model.

• Zip the contents of sample_code_submission (without the directory, but with metadata) to create a submission to the challenge.

• Alternatively, to create a sample result submission run:

  python ingestion_program/ingestion.py public_data sample_result_submission ingestion_program sample_code_submission

• Zip the contents of sample_result_submission (without the directory).

(2) If you are a challenge organizer and use this starting kit as a template, ensure that:

• you modify README.ipynb to provide a good introduction to the problem and good data visualization

• sample_data is a small data subset carved out the challenge TRAINING data, for practice purposes only (do not compromise real validation or test data)

• the following programs run properly

  python ingestion_program/ingestion.py sample_data sample_result_submission ingestion_program sample_code_submission

  python scoring_program/score.py sample_data sample_result_submission scoring_output

• IMPORTANT: if you switch between sample data, remove xxx_model.pickle from sample_code_submission, otherwise you'll have inconsistent data and models.

• the metric identified by metric.txt in the utilities directory is the metric used both to compute performances in README.ipynb and for the challenge. To use your own metric, change my_metric.py.