Deep learning

Course project of Deep Learning(11785)

1. Comparing performance of image semantic segmentation between RGB and RGBD dataset.
2. Test the performance of an existing method (U-Net) on large number class segmentation.
3. Test different fusion methods for integrating depth information
4. Improve the segmentation by fine tuning and changing number of classes.
5. Post processing by showing depth visual effects.

Course project of Machine learning for Signal Processing (18797)

Airbus challenge on ship detection from satellite imagery

1. Data augmentation on ship images: address the problem the images with ships are sparse
2. Unet baseline model implementation for ship position localization
3. Test different image augmentation methods for better performance
4. U-net with pretrained ResNet as encoder.

To be updated ...

1. CNN + LSTM for voice recognition.
2. LSTM for Shakespeare words generation.
3. Listen Attend and Spell for auto voice to words.
4. Autoenoder for simple image encoding