מחבר: Mihai Nan
Once a year, in the Disney Kingdom, the most spectacular event in the world of animation takes place:
the Disney Festival of Cartoons, where studios from all realms showcase their mastery in the art of imagery.
This year, however, the Grand Animator — guardian of visual harmony between worlds — faces an unexpected problem.
A sudden magical storm has caused the portals between universes to collide, mixing together scenes from four iconic realms:
The images from these worlds have been scattered everywhere, and now only a true Master of Visual Arts can distinguish them again.
The Grand Animator entrusts you with this mission:
“Young artist, the styles of these worlds are unique and sacred.
Help us identify them and restore harmony before the spell spreads further!”
You have two essential files, recovered directly from the magical portals:
train.jsonContains:
image_path – path to the imagecartoon_class – one of the 4 visual stylestest.jsonContains the same fields except for the class, which you must predict.
All images are stored in the images/ directory.
Build an image classification model capable of distinguishing the four visual styles:
You may use any modern architecture: ResNet, EfficientNet, Vision Transformer, ConvNeXt, etc., ideally with ImageNet pretraining and fine-tuning.
In the end, you must generate a submission.csv file with the format:
image_path,cartoon_class
images/test_0001.png,Tarzan
images/test_0002.png,Snow White
Your model's accuracy will determine how well you restored the balance across the four realms:
accuracy = (number_of_correct_predictions / total_number_of_predictions)
The colors, styles, and magic depend on you. Make the worlds shine together once more! 🌟🎨