Compilation of small projects during BMDATA courses:
This work use multi-class classification to with a dataset of clothes. The objective is to predict the class of an image between 8 classes.
The approached subjects are :
- Image Processing
- Creation and training of a classical Convolutionnal Neural Network
- Transfer Learning from pretrained VGG16 on ImageNet
- KNN to find similar images
- PCA to visualize image clusters from their features
As a result, built models were able to predict with an accuracy of ~90%.
The dataset is available at this link. The models weights and optimizer parameters are available at this link
I study autoencoders with the stl10 dataset. The objective is to predict the class of the images. The approached subjects are :
- Image Processing
- Basic Autoencoders
- Transfer Learning (from .npy files)
- Features Extraction The autoencoder has already been trained.
This work was about time series. The idea was to compare the best netowrks between CNN2D, CNN1D and LSTM with an application on movement recognition. The approached subjects are :
- Time series Processing
- Application of Conv2D, Conv1D and LSTM, and comparison of their performances.
- LSTM with pad and masking sequences
A Basic NN to predict the sentiment of a movie review. The subject approached are :
- Texts processing
- Creation of a dictionnary and one-hot encoding of texts
- Creation and Training of a MLP
- Visualisation of the predictions
Sentiment classification of a face. The subject approached are :
- Image processing
- Classification : MLP and CNN
- Video Capture on Colab
- Comparison between landmarks and face to predict sentiment