A Hybrid CPU-GPU System for Stitching Large Scale Optical Microscopy Images

Blattner, Timothy; Keyrouz, Walid; Chalfoun, Joe; Stivalet, Bertrand; Brady, Mary; Zhou, Shujia

doi:10.1109/icpp.2014.9

Cited by 26 publications

(14 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 1 compares the HTGS-based implementation of hybrid microscopy image stitching with the original implementation that did not use HTGS [6]. Each test case was repeated 50 times using a grid of 42 × 59 images (6.75 GB) and the average end-to-end run-time is reported.…”

Section: Microscopy Image Stitching With Htgsmentioning

confidence: 99%

“…This paper builds on our previous work [6] and formalizes our approach to developing a scalable implementation of image stitching for large optical microscopy images. It presents the Hybrid Task Graph Scheduler (HTGS), which is designed to aid in building hybrid workflows for high performance image processing.…”

Section: Introductionmentioning

confidence: 99%

“…Our work [6] uses a direct method, Kuglin and Hines’ phase correlation image alignment method [13], that is modified to use normalized correlation coefficients as described by Lewis [14, 15]. This method uses Fast Fourier Transforms (FFTs) to compute Fourier Correlation Coefficients and then uses these correlation coefficients to determine image displacements.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Hybrid Task Graph Scheduler for High Performance Image Processing Workflows

Blattner

Keyrouz

Bhattacharyya

et al. 2017

J Sign Process Syst

Self Cite

View full text Add to dashboard Cite

Designing applications for scalability is key to improving their performance in hybrid and cluster computing. Scheduling code to utilize parallelism is difficult, particularly when dealing with data dependencies, memory management, data motion, and processor occupancy. The Hybrid Task Graph Scheduler (HTGS) improves programmer productivity when implementing hybrid workflows for multi-core and multi-GPU systems. The Hybrid Task Graph Scheduler (HTGS) is an abstract execution model, framework, and API that increases programmer productivity when implementing hybrid workflows for such systems. HTGS manages dependencies between tasks, represents CPU and GPU memories independently, overlaps computations with disk I/O and memory transfers, keeps multiple GPUs occupied, and uses all available compute resources. Through these abstractions, data motion and memory are explicit; this makes data locality decisions more accessible. To demonstrate the HTGS application program interface (API), we present implementations of two example algorithms: (1) a matrix multiplication that shows how easily task graphs can be used; and (2) a hybrid implementation of microscopy image stitching that reduces code size by ≈ 43% compared to a manually coded hybrid workflow implementation and showcases the minimal overhead of task graphs in HTGS. Both of the HTGS-based implementations show good performance. In image stitching the HTGS implementation achieves similar performance to the hybrid workflow implementation. Matrix multiplication with HTGS achieves 1.3× and 1.8× speedup over the multi-threaded OpenBLAS library for 16k × 16k and 32k × 32k size matrices, respectively.

show abstract

Section: Microscopy Image Stitching With Htgsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Hybrid Task Graph Scheduler for High Performance Image Processing Workflows

Blattner

Keyrouz

Bhattacharyya

et al. 2017

J Sign Process Syst

Self Cite

View full text Add to dashboard Cite

show abstract

“…Micritic limestone (rock) CTE testing may have also presented problems since both in Figures 9a and 9b and in Figure 10c, it does not follow the expected trends. It is important to point out that the CTE is dependent on the testing temperature range [60] and since the aggregates were tested between 0 °C and 100 °C, while the cores were evaluated between 10 °C to 50 °C, a 1:1 correlation was not necessarily expected. Still, a generally good agreement is observed between the values obtained on the bulk cores and the small cylinders using the two different measurement techniques, considering the average CoV of 18 % for the NIST CTE data.…”

Section: Rock (Core) Propertiesmentioning

confidence: 99%

“…The four sets of images were then stitched together to create a single image using the ImageJ plugin MIST [60]. Using the silicon image as a reference, the MIST algorithm determined the translations and rotations necessary to stitch the image into a single image; the stitching was then executed for the other EDX images, as well as the BE image, using the same set of translations and rotations.…”

Section: Testing Performed At Nist (Sem Preparation and Analysis)mentioning

confidence: 99%

Influence of aggregate characteristics on concrete performance

Bentz¹,

Arnold²,

Boisclair³

et al. 2017

View full text Add to dashboard Cite

While the influence of paste properties on concrete performance has been extensively studied and in many cases reduced to quantitative relationships (e.g., Abram's law), that between aggregate characteristics and concrete performance has not been investigated in detail. Based on previous research that demonstrated significant strength differences for two similar concrete mixtures, one prepared with limestone aggregates and the other with siliceous gravel, a joint study between the National Institute of Standards and Technology (NIST) and the Federal Highway Administration (FHWA) was initiated to explore in detail the influence of aggregate source, mineralogy, and material properties on concrete performance. Eleven aggregates of differing mineralogy were identified and obtained both for bulk characterization and for incorporation into two concrete mixtures. The first concrete mixture was based on a 100 % ordinary Type I/II portland cement (OPC), while the second consisted of a ternary 60:30:10 volumetric blend of this cement with 30 % of a Class C fly ash and 10 % of a fine limestone powder. This latter sustainable mixture had exhibited exemplary performance in a previous study. Aggregates were characterized with respect to mechanical and thermomechanical properties, geometrical characteristics, and surface energies. For the prepared concretes, mechanical, thermomechanical, and electrical properties were measured at different ages out to 91 d and microstructural examinations were conducted to examine the interfaces between aggregates and cement paste. Concrete performance varied widely amongst the different aggregates, with the (range/average) ratio for 28-d compressive strength being 0.32 for the OPC concretes and 0.37 for those based on the ternary blend binder. With the exceptions of relating concrete modulus to aggregate modulus and concrete coefficient of thermal expansion (CTE) to aggregate CTE, weak correlations were generally obtained between a single aggregate characteristic and concrete performance properties. Models to predict 28-d compressive strength based on the aggregates' CTE (and aggregate absorption in the case of the ternary blend mixtures) provided predictions with a relative standard error (standard error/mean) of about 7 %. It is suggested that aggregate and binder characteristics control the bond between aggregates and paste. Then, for most properties, concrete performance is primarily controlled by the level of this bonding, a characteristic that was only assessed in an indirect manner in the present study. Research using non-linear ultrasonic measurements to better assess this bonding in specimens remaining from the present study is currently underway.

show abstract