Improving Fine-Grained Object Classification Using Adversarial Generated Unlabelled Samples

AbstractConvolutional neural networks (CNNs) trained for object recognition have been widely used to account for visually-driven neural responses in both the human and primate brains. However, because of the generality and complexity of the task of object classification, it is often difficult to make precise inferences about neural information processing using CNN representations from object classification despite the fact that these representations are effective for predicting brain activity. To better understand underlying the nature of the visual features encoded in different brain regions of the human brain, we predicted brain responses to images using fine-grained representations drawn from 19 specific computer vision tasks. Individual encoding models for each task were constructed and then applied to BOLD5000—a large-scale dataset comprised of fMRI scans collected while observers viewed over 5000 naturalistic scene and object images. Because different encoding models predict activity in different brain regions, we were able to associate specific vision tasks with each region. For example, within scene-selective brain regions, features from 3D tasks such as 3D keypoints and 3D edges explain greater variance as compared to 2D tasks—a pattern that replicates across the whole brain. Using results across all 19 task representations, we constructed a “task graph” based on the spatial layout of well-predicted brain areas from each task. We then compared the brain-derived task structure with the task structure derived from transfer learning accuracy in order to assess the degree of shared information between the two task spaces. These computationally-driven results—arising out of state-of-the-art computer vision methods—begin to reveal the task-specific architecture of the human visual system.

Download Full-text

Application of an image management system in microscopy

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100129899 ◽

1992 ◽

Vol 50 (2) ◽

pp. 1052-1053

Author(s):

Richard S. Chemock

Keyword(s):

Data Storage ◽

Cost Savings ◽

Image Data ◽

Analytical Techniques ◽

Operational Mode ◽

Fine Grained ◽

Image Recording ◽

Primary Output ◽

Capacity Data ◽

Image Management System

One of the most common tasks in a typical analysis lab is the recording of images. Many analytical techniques (TEM, SEM, and metallography for example) produce images as their primary output. Until recently, the most common method of recording images was by using film. Current PS/2R systems offer very large capacity data storage devices and high resolution displays, making it practical to work with analytical images on PS/2s, thereby sidestepping the traditional film and darkroom steps. This change in operational mode offers many benefits: cost savings, throughput, archiving and searching capabilities as well as direct incorporation of the image data into reports.The conventional way to record images involves film, either sheet film (with its associated wet chemistry) for TEM or PolaroidR film for SEM and light microscopy. Although film is inconvenient, it does have the highest quality of all available image recording techniques. The fine grained film used for TEM has a resolution that would exceed a 4096x4096x16 bit digital image.

Download Full-text

Use of fractals to describe microstructures of porous thick-film platinum electrodes

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100121971 ◽

1992 ◽

Vol 50 (1) ◽

pp. 314-315

Author(s):

Steven D. Toteda

Keyword(s):

Fractal Dimension ◽

Power Plants ◽

Electrical Response ◽

Cubic Phase ◽

Platinum Electrodes ◽

Fine Grained ◽

Impedance Response ◽

Platinum Particles ◽

Energy Surfaces ◽

Highly Porous

Zirconia oxygen sensors, in such applications as power plants and automobiles, generally utilize platinum electrodes for the catalytic reaction of dissociating O2 at the surface. The microstructure of the platinum electrode defines the resulting electrical response. The electrode must be porous enough to allow the oxygen to reach the zirconia surface while still remaining electrically continuous. At low sintering temperatures, the platinum is highly porous and fine grained. The platinum particles sinter together as the firing temperatures are increased. As the sintering temperatures are raised even further, the surface of the platinum begins to facet with lower energy surfaces. These microstructural changes can be seen in Figures 1 and 2, but the goal of the work is to characterize the microstructure by its fractal dimension and then relate the fractal dimension to the electrical response. The sensors were fabricated from zirconia powder stabilized in the cubic phase with 8 mol% percent yttria. Each substrate was sintered for 14 hours at 1200°C. The resulting zirconia pellets, 13mm in diameter and 2mm in thickness, were roughly 97 to 98 percent of theoretical density. The Engelhard #6082 platinum paste was applied to the zirconia disks after they were mechanically polished ( diamond). The electrodes were then sintered at temperatures ranging from 600°C to 1000°C. Each sensor was tested to determine the impedance response from 1Hz to 5,000Hz. These frequencies correspond to the electrode at the test temperature of 600°C.

Download Full-text