And it’s a touchdown! You’ve finally done.
You’re on the planet of Mars! After you’ve set foot on it, propped your flag, had your Neil Armstrong moment, you’re ready to click some pictures and share the data with your fellow scientists (and Instagram followers). Oops, the images your rover is sending are getting corrupted.
You must solve this image emergency by correcting these images. Click here for the starter kit.
In this challenge, you need to get multiple images, with black pixels, wrong orientations and such. These images need to be corrected and cleaned up. The corrected images together form the right image. The corrupted images can vary in size but the dimension of the final image is 512x512.
The single zip ( for example
train.zip ) contains two folders :-
Corrupted_Images the folder has multiple files in a tree form as shown below.
data ├── Corrupted_Images │ └── 0 │ ├── start_image.jpg │ ├── cioilbhbot.jpg │ ├── bcepylgtkz.jpg │ └── last_image.jpg └── Labels └── 0.jpg
Your model will take files from start_image.jpg, bcepylgtkz.jpg, cioilbhbot.jpg and last_image.jpg from
Corrupted_Images and return 0.jpg from
To ensure the right orientation of the output image, two files in corrupted samples will be labelled ( with right orientation & no corruption ), in the tree above. The file start_image.jpg indicates the top left image part of the image and last_image.jpg is the bottom right part of the image.
Following files are available in the
20000samples) This zip contains the training dataset containing the corrupted Images and Labels folder.
2000samples) This zip contains a validation dataset containing the corrupted Images and Labels folder.
5000samples) This zip will be used for actual evaluation for the leaderboard!
- Prepare a zip containing Labels folder containing predicted images with the name matching the corresponding folder names of
- The name of the above file should be submission.zip.
- Sample submission format available at sample_submission.zip in the resources section.
Make your first submission here 🚀 !!
🖊 Evaluation Criteria
During the evaluation, the average Mean Squared Error will be calculated over all the testing images.
np.mean((real_img - predicted_img)**2) is the code for calculating MSE for images.
- 💪 Challenge Page: https://www.aicrowd.com/challenges/image-correction
- 🗣️ Discussion Forum: https://www.aicrowd.com/challenges/image-correction/discussion
- 🏆 Leaderboard: https://www.aicrowd.com/challenges/image-correction/leaderboards
[Baseline] Image Correction