We present an extension to masked autoencoders (MAE) which improves on the representations learnt by the model by explicitly encouraging the learning of higher scene-level features. We do this by: (i) the introduction of a perceptual similarity term between generated and real images (ii) incorporating several techniques from the adversarial training literature including multi-scale training and adaptive discriminator augmentation. The combination of these results in not only better pixel reconstruction but also representations which appear to capture better higher-level details within images. More consequentially, we show how our method, Perceptual MAE, leads to better performance when used for downstream tasks outperforming previous methods. We achieve 78.1% top-1 accuracy linear probing on ImageNet-1K and up to 88.1% when fine-tuning, with similar results for other downstream tasks, all without use of additional pre-trained models or data.

Deep reinforcement learning (DRL) has the potential to train robots to perform complex tasks in the real world without requiring accurate models of the robot or its environment. However, agents trained with these algorithms typically lack the explainability of more traditional control methods. In this work, we use a combination of out-of-distribution generalisation tests and post hoc interpretability methods in order to understand what strategies DRL-trained agents use to perform a reaching task. To do so, we train agents under different conditions, using comparison to better interpret both quantitative and qualitative results; this allows us to not only provide local explanations, but also broad categorisations of behaviour. A key aim of our work is to understand how agents trained with visual domain randomisation (DR)—a technique which allows agents to generalise from simulation-based-training to the real world—differ from agents trained without. Our results show that the primary outcome of DR is more robust, entangled representations, accompanied by greater spatial structure in convolutional filters. Furthermore, even with an improved saliency method introduced in this work, we show that qualitative studies may not always correspond with quantitative measures, necessitating the combination of inspection tools in order to provide sufficient insights into the behaviour of trained agents. We conclude with recommendations for applying interpretability methods to DRL agents.

Among the greatest of the challenges of minimally invasive surgery (MIS) is the inadequate visualisation of the surgical field through keyhole incisions. Moreover, occlusions caused by instruments or bleeding can completely obfuscate anatomical landmarks, reduce surgical vision and lead to iatrogenic injury. The aim of this paper is to propose an unsupervised end-to-end deep learning framework, based on fully convolutional neural networks to reconstruct the view of the surgical scene under occlusions and provide the surgeon with intraoperative see-through vision in these areas. A novel generative densely connected encoder-decoder architecture has been designed which enables the incorporation of temporal information by introducing a new type of 3D convolution, the so called 3D partial convolution, to enhance the learning capabilities of the network and fuse temporal and spatial information. To train the proposed framework, a unique loss function has been proposed which combines feature matching, reconstruction, style, temporal and adversarial loss terms, for generating high fidelity image reconstructions. Advancing the state-of-the-art, our method can reconstruct the underlying view obstructed by irregularly shaped occlusions of divergent size, location and orientation. The proposed method has been validated on in vivo MIS video data, as well as natural scenes on a range of occlusion-to-image (OIR) ratios. It has also been compared against the latest video inpainting models in terms of image reconstruction quality using different assessment metrics. The performance evaluation analysis verifies the superiority of our proposed method and its potential clinical value.