Can we leverage high-resolution information without the unsustainable quadratic complexity to input scale? We propose Traversal Network (TNet), a novel multi-scale hard-attention architecture, which traverses image scale-space in a top-down fashion, visiting only the most informative image regions along the way. TNet offers an adjustable trade-off between accuracy and complexity, by changing the number of attended image locations. We compare our model against hard-attention baselines on ImageNet, achieving higher accuracy with less resources (FLOPs, processing time and memory). We further test our model on fMoW dataset, where we process satellite images of size up to $896 \times 896$ px, getting up to $2.5$x faster processing compared to baselines operating on the same resolution, while achieving higher accuracy as well. TNet is modular, meaning that most classification models could be adopted as its backbone for feature extraction, making the reported performance gains orthogonal to benefits offered by existing optimized deep models. Finally, hard-attention guarantees a degree of interpretability to our model's predictions, without any extra cost beyond inference.

We address the problem of shoulder-surfing attacks on authentication schemes by proposing IllusionPIN (IPIN), a PIN-based authentication method that operates on touchscreen devices. IPIN uses the technique of hybrid images to blend two keypads with different digit orderings in such a way, that the user who is close to the device is seeing one keypad to enter her PIN, while the attacker who is looking at the device from a bigger distance is seeing only the other keypad. The user's keypad is shuffled in every authentication attempt, since the attacker may memorize the spatial arrangement of the pressed digits. To reason about the security of IPIN, we developed an algorithm which is based on human visual perception and estimates the minimum distance from which an observer is unable to interpret the keypad of the user. We tested our estimations with 84 simulated shoulder-surfing attacks from 21 different people. None of the attacks was successful against our estimations. In addition, we estimated the minimum distance from which a camera is unable to capture the visual information from the keypad of the user. Based on our analysis, it seems practically almost impossible for a surveillance camera to capture the PIN of a smartphone user when IPIN is in use.