ImgLblReader

The ImgLblReader class is used to load images from various paths into NumPy arrays and parsing their labels. Currently, the ImgLblReader class creates an iterator to iterate over all images in a random order and shuffles its internal order after visiting each image exactly once. Future release will cover more features.

At the moment, ImgLblReader only parses ball annotations in ‘AI Ball’ format of the Bit-Bots ImageTagger.

Instantiate

dat = ImgLblReader(pathlist: List[str],
                   batch_size: int = 1,
                   queue_size: int = 16,
                   processes: int = None,
                   img_dim: tuple = (200, 150),
                   filter_labels = False)

• pathlist: a list of strings describing each path that ImgReader should check for images
• batch_size: images are automatically stored in batches with an alterable batch size
• queue_size: size of the internal multiprocessing queue; this setting affects memory usage
• processes: by default half of the available CPU cores are used, but you can define any number of worker processes manually
• img_dim: (width, height) to resize each image to
• filter_labels: deprecated; will be removed in a future release

Methods

• get_dataset_size(): returns the total number of images in all given paths
• get_max_queue_size(): returns the queue’s capacity
• get_next_batch(): pops the next ImageBatch object from queue; blocks automatically if queue is empty; each ImageBatch contains the original images and the corresponding labels