Malte Lüken

It seems that the pretrained AU models expect different input sizes for the landmarks argument. A detector with `au_model='JAANET'` allows for a landmarks argument with 3 dimensions (e.g., landmarks for...

Replace whisper with faster-whisper

1

Replaces whisper with faster-whisper to speed up the transcription. Removes ffmpeg as a requirement. Adds a new argument `compute_type` to `AudioTranscriber.transcribe()`.

The models for detecting faces and computing face representations could be replaced by more SOTA ones, e.g.,: - RetinaFace instead of MTCNN (see https://github.com/biubug6/Pytorch_Retinaface?tab=readme-ov-file) - FRL (see https://github.com/1adrianb/unsupervised-face-representation) To include...

We currently disabled doctest in our documentation test, since our examples take a lot of computational ressources to run. In the future, however, we should find a way to test...

When giving a tf.keras.metrics object as metric in `find_best_architecture()`, e.g., `find_best_architecture(..., metric=tf.keras.metrics.Precision())`, an error is raised: ``` models = modelgen.generate_models(X_train.shape, y_train.shape[1], number_of_models=number_of_models, task=task, metrics=[metric], **kwargs) _, val_performance, _ = train_models_on_samples(X_train,...

Applying the template to existing code with tests creates dummy test files

2

The template creates the dummy test files `tests/test_my_module.py`, `tests/test_values.py`, and `tests/test_projects.py` when I apply it to code which already has tests in a `tests/` folder. This is not really necessary...