Vision language models (VLMs) that enable natural language interaction with satellite imagery can democratize Earth observation by accelerating expert workflows, making data accessible to non-specialists, and enabling planet-scale automation. However, existing datasets focus mainly on short-term, high-resolution imagery from a limited number of satellites, overlooking low-resolution, multi-satellite, long-term archives, such as Landsat, that are essential for affordable and bias-robust global monitoring. We address this gap with Landsat30-AU, a large-scale vision-language dataset built from 30-meter resolution imagery collected by four Landsat satellites (5, 7, 8, and 9) over Australia, spanning more than 36 years. The dataset includes two components: Landsat30-AU-Cap, containing 196,262 image-caption pairs, and Landsat30-AU-VQA, comprising 17,725 human-verified visual question answering (VQA) samples across eight remote sensing domains. Both datasets are curated through a bootstrapped pipeline that leverages generic VLMs with iterative refinement and human verification to ensure quality. Our evaluation of eight VLMs on our benchmark reveals that off-the-shelf models struggle to understand satellite imagery. The open-source remote-sensing VLM EarthDial achieves only 0.07 SPIDEr in captioning and a VQA accuracy of 0.48, highlighting the limitations of current approaches. Encouragingly, lightweight fine-tuning of Qwen2.5-VL-7B on Landsat30-AU improves captioning performance from 0.11 to 0.31 SPIDEr and boosts VQA accuracy from 0.74 to 0.87.

Cloud computing services offer highly reliable, scalable and efficient solutions with a large pool of easily accessible, virtualized resources. They are becoming an increasingly prevalent delivery model. We have developed a cloud-based image analysis toolbox to provide a wide user base with easy access to the software tools we have developed over the last decade. The toolbox is provided as a service on an Australian national cloud infrastructure. The design and implementation of the cloud-based service are presented, including its architecture, key components and some image analysis and visualization examples showing the capabilities of the service for biomedical image analysis.

Computed Tomography (CT) is a non-destructive imaging technique widely used across many scientific, industrial and medical fields. It is both computationally and data intensive, and therefore can benefit from infrastructure in the “supercomputing” domain for research purposes, such as Synchrotron science. Our group within CSIRO has been actively developing X-ray tomography and image processing software and systems for HPC clusters. We have also leveraged the use of GPU's (Graphical Processing Units) for several codes enabling speedups by an order of magnitude or more over CPU-only implementations. A key goal of our systems is to enable our targeted “end users”, researchers, easy access to the tools, computational resources and data via familiar interfaces and client applications such that specialized HPC expertise and support is generally not required in order to initiate and control data processing, analysis and visualzation workflows. We have strived to enable the use of HPC facilities in an interactive fashion, similar to the familiar Windows desktop environment, in contrast to the traditional batch-job oriented environment that is still the norm at most HPC installations. Several collaborations have been formed, and we currently have our systems deployed on two clusters within CSIRO, Australia. A major installation at the Australian Synchrotron (MASSIVE GPU cluster) where the system has been integrated with the Imaging and Medical Beamline (IMBL) detector to provide rapid on-demand CT-reconstruction and visualization capabilities to researchers whilst on-site and remotely. A smaller-scale installation has also been deployed on a mini-cluster at the Shanghai Synchrotron Radiation Facility (SSRF) in China. All clusters run the Windows HPC Server 2008 R2 operating system. The two large clusters running our software, MASSIVE and CSIRO Bragg are currently configured as “hybrid clusters” in which individual nodes can be dual-booted between Linux and Windows as demand requires. We have also recently explored the adaptation of our CT-reconstruction code to Cloud infrastructure, and have constructed a working “proof-of-concept” system for the Microsoft Azure Cloud. However, at this stage several challenges remain to be met in order to make it a truly viable alternative to our HPC cluster solution. Recently, CSIRO was successful in its proposal to develop eResearch tools for the Australian Government funded NeCTAR Research Cloud. As part of this project our group will be contributing CT and imaging processing components.

Efficiently and effectively identifying similar videos is an important and nontrivial problem in content-based video retrieval. This paper proposes a subspace symbolization approach, namely SUDS, for content-based retrieval on very large video databases. The novelty of SUDS is that it explores the data distribution in subspaces to build a visual dictionary with which the videos are processed by deriving the string matching techniques with two-step data simplification. Specifically, we first propose an adaptive approach, called VLP, to extract a series of dominant subspaces of variable lengths from the whole visual feature space without the constraint of dimension consecutiveness. A stable visual dictionary is built by clustering the video keyframes over each dominant subspace. A compact video representation model is developed by transforming each keyframe into a word that is a series of symbols in the dominant subspaces, and further each video into a series of words. Then, we present an innovative similarity measure called CVE, which adopts a complementary information compensation scheme based on the visual features and sequence context of videos. Finally, an efficient two-layered index strategy with a number of query optimizations is proposed to facilitate video retrieval. The experimental results demonstrate the high effectiveness and efficiency of SUDS.

We propose a subspace symbolization approach, namely SUDS, for content-based search on very large video databases. The novelty of SUDS is that it explores the data distribution in subspaces to build a visual dictionary. With this dictionary, the video data are processed using string matching techniques with two-step data simplification. A compact video representation model is developed by transforming each keyframe into a word that is a series of symbols in the dominant subspaces. Then, we present an innovative similarity measure called ED, which draws from the concept of the edit distance on strings to conduct video matching. The experimental results demonstrate the high effectiveness of SUDS with optimal parameters.

We propose a shot-based interest point selection approach for effective and efficient near-duplicate search over a large collection of video shots. The basic idea is to eliminate the local descriptors with lower frequencies among the selected video frames from a shot to ensure that the shot representation is compact and discriminative. Specifically, we propose an adaptive frame selection strategy called furthest point voronoi (FPV) to produce the shot frame set according to the shot content and frame distribution. We describe a novel strategy named reference extraction (RE) to extract the shot interest descriptors from a keyframe with the support of the selected frame set. We demonstrate the effectiveness and efficiency of the proposed approaches with extensive experiments.

It is noted that relatively few commercial practitioners have embraced Ada, chiefly because of the lack of efficient compilers and the high initial cost of using it. The results of efforts to address these concerns are summarized. Several Ada compilers targeted to different commercial microprocessors have been benchmarked. This benchmarking activity included obtaining the execution time as well as the memory requirements of Ada compilers and comparing them with compilers for traditional languages like Pascal. In order to determine the cost impact of using Ada, a model has been developed from first principles. The model predicts the cost of using Ada compared to other high languages encompasses the developmental phase as well as the maintenance phase. The model is further refined to include the impact of reusing software. On the basis of these data, it is concluded that Ada technology has matured enough to be used in the embedded environment and can represent a potential cost savings compared to other languages over the entire life cycle.>