This paper introduces two extensions to the popular PyTorch machine learning framework, TorchDynamo and TorchInductor, which implement the torch.compile feature released in PyTorch 2. TorchDynamo is a Python-level just-in-time (JIT) compiler that enables graph compilation in PyTorch programs without sacrificing the flexibility of Python. It achieves this by dynamically modifying Python bytecode before execution and extracting sequences of PyTorch operations into an FX graph, which is then JIT compiled using one of many extensible backends. TorchInductor is the default compiler backend for TorchDynamo, which translates PyTorch programs into OpenAI's Triton for GPUs and C++ for CPUs. Results show that TorchDynamo is able to capture graphs more robustly than prior approaches while adding minimal overhead, and TorchInductor is able to provide a 2.27× inference and 1.41× training geometric mean speedup on an NVIDIA A100 GPU across 180+ real-world models, which outperforms six other compilers. These extensions provide a new way to apply optimizations through compilers in eager mode frameworks like PyTorch.

A wide variety of computer vision applications rely on superpixel or supervoxel algorithms as a preprocessing step. This underlines the overall importance that these approaches have gained in recent years. However, most methods show a lack of temporal consistency or fail in producing temporally stable superpixels. In this paper, we present an approach to generate temporally consistent superpixels for video content. Our method is formulated as a contour-evolving expectation-maximization framework, which utilizes an efficient label propagation scheme to encourage the preservation of superpixel shapes and their relative positioning over time. By explicitly detecting the occlusion of superpixels and the disocclusion of new image regions, our framework is able to terminate and create superpixels whose corresponding image region becomes hidden or newly appears. Additionally, the occluded parts of superpixels are incorporated in the further optimization. This increases the compliance of the superpixel flow with the optical flow present in the scene. Using established benchmark suites, we show the performance of our approach in comparison to state-of-the-art supervoxel and superpixel algorithms for video content.

Many recent superpixel algorithms for video content rely on dense optical flow vectors to propagate segmentation results from one frame to the next. In this paper, we assess the impact of the optical flow quality on the over-segmentation quality. Our evaluation shows that it is indispensable for videos with large object displacement and camera motion. But due to the high computational costs high-quality, dense optical flow is not suitable for real-time applications. Therefore, we propose a fast propagation scheme that is based on sparse feature tracking and mesh-based image warping. In a thorough evaluation, we compare our proposed scheme to the results of other state-of-the-art propagation methods using established benchmarks. The results show that our method speeds up the propagation process by a factor of 100 while producing a comparable segmentation quality.

A major constraint for mobile surveillance systems (e.g. UAV-mounted) is the limited available bandwidth. Thus, ROI-based video coding, where only important image parts are transmitted with high data rate, seems to be a suitable solution for these applications. The challenging part in these systems is to reliably detect the regions of interest (ROI) and to select the corresponding macroblocks to be encoded in high quality. Recently, difference image-based moving object detectors have been used for this purpose. But as these pixel-based approaches fail for homogeneous parts of moving objects, which leads to artifacts in the decoded video stream, this paper proposes the use of a superpixel-based macroblock selection. It is shown that the proposed approach significantly reduces artifacts by raising the moving objects detection rate from 7 % to over 96 % while increasing the data rate by only 20% to 1.5-4Mbit/s (full HD resolution, 30 Hz) still meeting the given constraints. Moreover, other simple approaches (dilation) as well as state-of-the-art GraphCut-based approaches are outperformed.

Super pixel algorithms represent a very useful and increasingly popular preprocessing step for a wide range of computer vision applications, as they offer the potential to boost efficiency and effectiveness. In this regards, this paper presents a highly competitive approach for temporally consistent super pixels for video content. The approach is based on energy-minimizing clustering utilizing a novel hybrid clustering strategy for a multi-dimensional feature space working in a global color subspace and local spatial subspaces. Moreover, a new contour evolution based strategy is introduced to ensure spatial coherency of the generated super pixels. For a thorough evaluation the proposed approach is compared to state of the art super voxel algorithms using established benchmarks and shows a superior performance.

Current Moving Object Detectors in airborne Region of Interest (ROI) coding systems for police surveillance applications used on-board of UAVs are often based on Global Motion Estimation (GME) techniques. Since in these scenarios the camera is moving, simple background removal approaches cannot be applied without a Global Motion Compensation (GMC). Common GMC algorithms assume the ground to be planar, allowing the pixels of the previous frame to be motion compensated into the current frame by applying a projective transformation. The difference image between the compensated frame and the current frame emphasis regions containing possible motion. Such moving object detectors are great in terms of run-time efficiency but are known to lack in terms of accuracy - especially for unstructured regions of moving objects - as well as the robustness against noise. Superpixel segmentation was recently proposed to overcome the issue of the imprecise region cuts given by the difference image. It provides a greatly improved true positive detection rate, but unintentionally also increases the area of false positives. This paper proposes the use of a mesh-based GME and GMC to detect the moving object regions wherein a cluster filter eliminates errors in the optical flow by assuming a smooth vector field as the global motion model. In doing so we improve the coding efficiency of the fully automatic ROI coding system by more than 24% for moving object areas conserving the detection benefits of the integration of superpixel segmentation.