🏆 Achievements:
- 🥈 2nd Place in Public Leaderboard
- 🥉 3rd Place in Private Leaderboard
This repository presents two different approaches to tackle a multi-label image classification problem using Futurama frames as the dataset.
This project was part of the Futurama Kaggle Competition hosted by Mediavida, focusing on classifying characters in frames from the show.
Two distinct methodologies were used:
- A custom Keras CNN model for character identification.
- A ResNet152 model pretrained with the Imaginet_v2 dataset.
Both approaches achieved competitive results, demonstrating robust performance in multi-label image classification tasks.
- Objective: Identify characters in Futurama frames using a custom convolutional neural network built with Keras.
- Competition Score:
- MCRMSE: 0.12527
- Notebook: Keras Conv Notebook
- Objective: Utilize the ResNet152 architecture pretrained on Imaginet_v2 to identify characters in Futurama frames.
- Competition Score:
- MCRMSE: 0.09149
- Notebook: ResNet152 Notebook
- Competition Page: Futurama Kaggle Competition
- Public Leaderboard: View Public Scores 🥈
- Private Leaderboard: View Private Scores 🥉
- Python 3.x
- Keras & TensorFlow for the custom CNN approach.
- PyTorch & torchvision for the ResNet152 approach.
- Other Libraries:
numpypandasmatplotlibseaborn
The project achieved excellent results on both the public and private leaderboards:
- Public Leaderboard: 2nd Place 🥈
- Private Leaderboard: 3rd Place 🥉
Scores were measured using Mean Columnwise Root Mean Square Error (MCRMSE):
- Keras Conv: 0.12527
- ResNet152: 0.09149
- Community: Mediavida Dev Forum
This project is licensed under the MIT License. See the LICENSE file for more details.