Type Definition vaal_sys::VAALContext

pub type VAALContext = vaal_context;

@struct VAALContext

Context object for VAAL to manage internal model data structures. The VAALContext manages the DeepViewRT NNContext as well as parameters related to model post-processing.

Models typically require further post-processing before the outputs are ready to be interpreted. What post-processing is required varies greatly between various model architectures. A common post-processing step is the decoder which is responsible for translating an internal model representation into one which represents the problem space.

For detection models the post-processing includes the box decoding step which can either be integrated into the model or run as a separate step using the model’s outputs followed by sorting and non-maximum supression “NMS” which is tasked with cleaning the bounding box output.

For classification models the post-processing is typically implemented by finding within an output vector the index of the maximum value “argmax” where the index has meaning. For example, the index into the labels with which the classifier was trained. The value itself can also be interpreted as a weak confidence metric in cases where a softmax or sigmoid was applied to the vector.