hcai-datasets 0.1.0

!Alpha Version! - This repository contains the backend server for the nova annotation ui (https://github.com/hcmlab/nova)

Description

This repository contains code to make datasets stored on th corpora network drive of the chair compatible with the tensorflow dataset api .

Currently available Datasets

Dataset	Status	Url
ckplus	✅	http://www.iainm.com/publications/Lucey2010-The-Extended/paper.pdf
affectnet	✅	http://mohammadmahoor.com/affectnet/
faces	✅	https://faces.mpdl.mpg.de/imeji/
nova_dynamic	✅	https://github.com/hcmlab/nova
audioset	❌	https://research.google.com/audioset/
is2021_ess	❌	-
librispeech	❌	https://www.openslr.org/12

Example Usage

import os
import tensorflow as tf
import tensorflow_datasets as tfds
import hcai_datasets
from matplotlib import pyplot as plt

# Preprocessing function
def preprocess(x, y):
  img = x.numpy()
  return img, y

# Creating a dataset
ds, ds_info = tfds.load(
  'hcai_example_dataset',
  split='train',
  with_info=True,
  as_supervised=True,
  builder_kwargs={'dataset_dir': os.path.join('path', 'to', 'directory')}
)

# Input output mapping
ds = ds.map(lambda x, y: (tf.py_function(func=preprocess, inp=[x, y], Tout=[tf.float32, tf.int64])))

# Manually iterate over dataset
img, label = next(ds.as_numpy_iterator())

# Visualize
plt.imshow(img / 255.)
plt.show()

Example Usage Nova Dynamic Data

import os
import hcai_datasets
import tensorflow_datasets as tfds
from sklearn.svm import LinearSVC
import numpy as np
from sklearn.calibration import CalibratedClassifierCV
import warnings
warnings.simplefilter("ignore")

## Load Data
ds, ds_info = tfds.load(
  'hcai_nova_dynamic',
  split='dynamic_split',
  with_info=True,
  as_supervised=True,
  data_dir='.',
  read_config=tfds.ReadConfig(
    shuffle_seed=1337
  ),
  builder_kwargs={
    # Database Config
    'db_config_path': 'nova_db.cfg',
    'db_config_dict': None,

    # Dataset Config
    'dataset': '<dataset_name>',
    'nova_data_dir': os.path.join('C:', 'Nova', 'Data'),
    'sessions': ['<session_name>'],
    'roles': ['<role_one>', '<role_two>'],
    'schemes': ['<label_scheme_one'],
    'annotator': '<annotator_id>',
    'data_streams': ['<stream_name>'],

    # Sample Config
    'frame_step': 1,
    'left_context': 0,
    'right_context': 0,
    'start': None,
    'end': None,
    'flatten_samples': False, 
    'supervised_keys': ['<role_one>.<stream_name>', '<scheme_two>'],

    # Additional Config
    'clear_cache' : True
  }
)

data_it = ds.as_numpy_iterator()
data_list = list(data_it)
data_list.sort(key=lambda x: int(x['frame'].decode('utf-8').split('_')[0]))
x = [v['<stream_name>'] for v in data_list]
y = [v['<scheme_two'] for v in data_list]

x_np = np.ma.concatenate( x, axis=0 )
y_np = np.array( y )

linear_svc = LinearSVC()
model = CalibratedClassifierCV(linear_svc,
                               method='sigmoid',
                               cv=3)
print('train_x shape: {} | train_x[0] shape: {}'.format(x_np.shape, x_np[0].shape))
model.fit(x_np, y_np)