Verification of Size Invariance in DNN Activations using Concept Embeddings
This work provides incremental insights into DNN interpretability for safety-critical domains like medical imaging and automated driving, focusing on sub-object concepts such as body parts.
The paper tackled the problem of understanding internal representations in deep neural networks for safety-critical applications by applying concept analysis to large object detectors, specifically assessing size invariance of body part representations, and found that these representations are mostly size invariant across three standard networks including Mask R-CNN.
The benefits of deep neural networks (DNNs) have become of interest for safety critical applications like medical ones or automated driving. Here, however, quantitative insights into the DNN inner representations are mandatory. One approach to this is concept analysis, which aims to establish a mapping between the internal representation of a DNN and intuitive semantic concepts. Such can be sub-objects like human body parts that are valuable for validation of pedestrian detection. To our knowledge, concept analysis has not yet been applied to large object detectors, specifically not for sub-parts. Therefore, this work first suggests a substantially improved version of the Net2Vec approach (arXiv:1801.03454) for post-hoc segmentation of sub-objects. Its practical applicability is then demonstrated on a new concept dataset by two exemplary assessments of three standard networks, including the larger Mask R-CNN model (arXiv:1703.06870): (1) the consistency of body part similarity, and (2) the invariance of internal representations of body parts with respect to the size in pixels of the depicted person. The findings show that the representation of body parts is mostly size invariant, which may suggest an early intelligent fusion of information in different size categories.