AI Detector Reviews

AI Detector Reviews — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Radar geo-warping

  Radar geo-warping is the adjustment of geo-referenced radar images and video data to be consistent with a geographical projection. This image warping avoids any restrictions when displaying it together with video from multiple radar sources or with other geographical data including scanned maps and satellite images which may be provided in a particular projection. There are many areas where geo warping has unique benefits: Single radar video signal displayed together with maps of different geographical projections. E.g. Mercator UTM stereographic Multiple radar video signals displayed simultaneously: Having the computing power to do so on one computer. Adapting the projection of all radar signals allowing the geographically correct display and accurate superimposition of those videos. Slant range correction: a modern 3D radar system can measure the height of a target and hence it is possible to correct the radar video by the real corrected range of the target. Slant Range Correction also allows to compensate the radar tower height e.g. for maritime surveillance radars. == Introduction == Radar video presents the echoes of electromagnetic waves a radar system has emitted and received as reflections afterwards. These echoes are typically presented on a computer screen with a color-coding scheme depicting the reflection strength. Two problems have to be solved during such a visualization process. The first problem arises from the fact that typically the radar antenna turns around its position and measures the reflection echo distances from its position in one direction. This effectively means that the radar video data are present in polar coordinates. In older systems the polar oriented picture has been displayed in so called plan position indicators (PPI). The PPI-scope uses a radial sweep pivoting about the center of the presentation. This results in a map-like picture of the area covered by the radar beam. A long-persistence screen is used so that the display remains visible until the sweep passes again. Bearing to the target is indicated by the target's angular position in relation to an imaginary line extending vertically from the sweep origin to the top of the scope. The top of the scope is either true north (when the indicator is operated in the true bearing mode) or ship's heading (when the indicator is operated in the relative bearing mode). For visualization on a modern computer screen the polar coordinates have to be converted into Cartesian coordinates. This process called radar scan conversion is presented with more detail in the next section. The second problem to solve arises from the fact that a radar system is placed in the real world and measures real world echo positions. These echoes have to be displayed together with other real world data like object positions, vector maps and satellite images in a consistent way. All this information refers to the curved earth surface but is displayed on a flat computer display. Building a link from real world earth positions to display pixels is commonly called geographical referencing or in short geo-referencing. Part of the geo-referencing process is to map the 3D earth surface onto a 2D display. This process of a geographical projection can be performed in many ways, but different data sources have their own 'natural' projection. E.g. Cartesian radar video data from a radar source on the earth surface are geo-referenced by a so-called radar projection. When using this radar projection the Cartesian radar video pixels can directly displayed on a computer screen (only being linearly transformed according to the current position on the screen and e.g. the current zoom level). A problem now arises if e.g. also a satellite map shall be shown together with the radar video data. The 'natural' geographical projection of a satellite image would be a satellite projection which depends on the satellite orbit, position and further parameters. Now either the satellite image has to be reprojected to a radar projection or the radar video has to use the satellite projection. This geographical re-projection is also called geographical warping or Geo Warping where each image pixel has to be transformed from one projection into another. This article describes in further detail the Geo Warping of radar video images in real time. It will also show that radar video Geo Warping is done most efficiently when it is integrated with the radar scan conversion process. == Radar-scan conversion == This section describes the principles of the radar-scan conversion (RSC) process. The radar supplies its measured data in polar coordinates (ρ,θ) directly from the rotating antenna. ρ defines the target/echo distance and θ the target angle in polar world coordinates. These data are measured, digitized and stored in a polar coordinate polar store or polar pixmap. The main RSC task is to convert these data to Cartesian (x, y) display coordinates, creating the necessary display pixels. The RSC process is influenced by the current zoom, shift and rotation settings defining which part of the 'world' shall be visible in the display image. As detailed later the RSC process also takes the currently used geographical projection into account when the radar video images are Geo Warped. The OpenGL RSC is implemented using a reverse scan conversion approach which calculates for every image pixel the most appropriate radar amplitude value in the polar store. This approach generates an optimal image without any artifacts known from forward spoke fill algorithms. By applying bi-linear filtering between adjacent pixels in the polar store during the conversion process the OpenGL RSC finally achieves a very high visual quality radar display image for every zoom level, creating smooth images of the radar echoes. == Radar projection == This section illustrates how radar video data are geo referenced and displayed on a computer screen. The radar sensor is positioned on the earth surface with a height h above the ground. It measures the direct distance d to the target (and not e.g. the distance the target is away from the radar if one would move on the earth surface). This distance is then used in the display plane after adjustment to the current display zoom level by the radar scan converter (RSC). Now it has to be clarified how the radar video data is geo referenced. This basically means, that if we want to display a geographical real world object (like e.g. a light house) which is at the same real world position as the radar target, that it also shall appear at the same position in the display plane. This is realized by calculating the distance from the radar sensor to the respective real world object and use that distance in the display plane. The position of the real world object is typically given in geographical coordinates (latitude, longitude and height above the earth surface). In other words, using a radar projection with geographical data is done by simulating a radar measurement process with the real world objects and use the resulting range and azimuth in the display plane. The second picture to the right shows an example radar projection with the center of projection (COP) at latitude 50.0° and longitude 0.0° which is also the radar position. The dashed lines are the equal-latitude and equal-longitude lines on top of the background map. The solid lines show equal-range and equal-azimuth with the respect to the radar position. It is a feature of the radar projection that equal-range lines are circles and equal-azimuth lines are straight lines. This is necessary to display radar video consistently with other map data when using a radar projection where the projection center has to be the radar position. == Geo Warping process == This section explains the actual geo warping or re-projection process when applied to radar video in real time. Assume we want to display radar video on top of a satellite image. As an example we use the CIB projection which is used to display satellite data in CIB (Controlled Image Base) format. The Figure Geo Warping Radar to CIB Projection shows dashed the maximal range circle for a range of 111 km or 60 miles using the radar projection. Such a range is typical for long range coastal surveillance radars. As stated in the last section this is a perfect circle also on the computer screen. The solid line ellipse shows the same range circle for the CIB projection. Typically the errors occurring without Geo Warping are smallest near the radar position if at least the projection center (COP) coincides with the radar position, as realized in our example. Otherwise the error distribution depends both on the used projection and also on the projection parameters. Thus, in our case the errors are most significant near the maximum radar range. The CIB projection error corrected in east–west direction at half the radar range is 2.6 km and is 5.3 km at the full radar range of 111 km. An error of 5.3 km is

  Read more →
• 

  Speculative decoding

  Speculative decoding is an inference-time optimization for autoregressive large language models (LLMs) that generates multiple tokens per decoding step instead of one. A smaller draft model proposes a sequence of candidate tokens, and the larger target model verifies them in a single forward pass through a modified rejection sampling scheme. The verification preserves the target model's original output distribution, so the technique produces the same results as standard decoding while cutting latency by roughly two to three times. The name is an analogy to speculative execution in CPU design, where a processor runs instructions along a predicted branch before the outcome is known. == Background == Standard autoregressive decoding in large language models generates one token at a time. The model computes a probability distribution over its vocabulary, samples the next token, and feeds that token back as input. For large models, this process is bottlenecked by memory bandwidth rather than arithmetic throughput: loading the model's parameters from high-bandwidth memory (HBM) to the processor takes up most of the wall-clock time at each step. Because of this, a forward pass over one token and a forward pass over several tokens in a batch take roughly the same time. Speculative decoding relies on this property. == Mechanism == The technique alternates between two phases: drafting and verification. During drafting, a fast approximation model generates a short run of K candidate tokens, typically between 3 and 12. The draft model is usually a much smaller version of the target model or a lightweight auxiliary network. During verification, the target model scores the entire draft sequence in one batched forward pass. A modified rejection sampling algorithm compares the draft and target probabilities at each position. If the target model would have been at least as likely to produce a given token, that token is accepted; the first token that fails is resampled from a corrected distribution, and everything after it is thrown out. The result is that the output distribution is the same as if each token had been generated one at a time. How many tokens get accepted per cycle depends on how well the draft model matches the target. For common words and predictable continuations the match tends to be good, so the target model can confirm several tokens at once. == History == An early precursor was blockwise parallel decoding, proposed in 2018 by Stern, Shazeer, and Uszkoreit. Their method predicted multiple future tokens through auxiliary prediction heads and validated them against the autoregressive model, but it only worked with greedy decoding and did not preserve the full sampling distribution. The modern form of the technique came from Yaniv Leviathan, Matan Kalman, and Yossi Matias at Google Research, who posted "Fast Inference from Transformers via Speculative Decoding" on arXiv in November 2022. Separately and at about the same time, Charlie Chen and colleagues at DeepMind arrived at a closely related method they called speculative sampling, published in February 2023. Both papers introduced the use of rejection sampling to guarantee that the output distribution is unchanged. Leviathan et al. showed roughly 2–3x speedup on T5-XXL (11 billion parameters); Chen et al. reported 2–2.5x on the Chinchilla model (70 billion parameters). The Leviathan et al. paper was presented as an oral at the International Conference on Machine Learning in July 2023. == Variants == SpecInfer (Miao et al., 2024) uses multiple small language models to jointly build a tree of candidate continuations rather than a single chain. The target model verifies the whole tree in parallel and keeps the longest valid path, with reported speedups of 1.5–3.5x. Medusa (Cai et al., 2024) takes a different approach by not using a separate draft model at all. Extra lightweight decoding heads are attached to the target model itself, and each one predicts a token at a different future position. The candidates are evaluated through a tree-structured attention mechanism. The authors measured 2.2–3.6x speedup. EAGLE (Li et al., 2024) performs autoregression on the target model's internal feature representations (specifically the second-to-top layer) rather than on tokens directly. On LLaMA 2 Chat 70B, this gave a 2.7–3.5x latency reduction. Later versions added dynamic draft trees (EAGLE-2) and further optimizations (EAGLE-3), reaching 3–6.5x speedup. == Adoption == By 2024, speculative decoding had become a standard part of production LLM serving. Google uses it in the AI Overviews feature of Google Search. Open-source inference frameworks such as vLLM, NVIDIA's TensorRT-LLM, and SGLang all include built-in support for speculative decoding and its variants. Apple, AWS, and Meta have also published research extending the method or deploying it at scale.

  Read more →
• 

  PatchMatch

  PatchMatch is an algorithm used to quickly find correspondences (or matches) between small square regions (or patches) of an image. It has various applications in image editing, such as reshuffling or removing objects from images or altering their aspect ratios without cropping or noticeably stretching them. PatchMatch was first presented in a 2011 paper by researchers at Princeton University. == Algorithm == The goal of the algorithm is to find the patch correspondence by defining a nearest-neighbor field (NNF) as a function f : R 2 → R 2 {\displaystyle f:\mathbb {R} ^{2}\to \mathbb {R} ^{2}} of offsets, which is over all possible matches of patch (location of patch centers) in image A, for some distance function of two patches D {\displaystyle D} . So, for a given patch coordinate a {\displaystyle a} in image A {\displaystyle A} and its corresponding nearest neighbor b {\displaystyle b} in image B {\displaystyle B} , f ( a ) {\displaystyle f(a)} is simply b − a {\displaystyle b-a} . However, if we search for every point in image B {\displaystyle B} , the work will be too hard to complete. So the following algorithm is done in a randomized approach in order to accelerate the calculation speed. The algorithm has three main components. Initially, the nearest-neighbor field is filled with either random offsets or some prior information. Next, an iterative update process is applied to the NNF, in which good patch offsets are propagated to adjacent pixels, followed by random search in the neighborhood of the best offset found so far. Independent of these three components, the algorithm also uses a coarse-to-fine approach by building an image pyramid to obtain the better result. === Initialization === When initializing with random offsets, we use independent uniform samples across the full range of image B {\displaystyle B} . This algorithm avoids using an initial guess from the previous level of the pyramid because in this way the algorithm can avoid being trapped in local minima. === Iteration === After initialization, the algorithm attempted to perform iterative process of improving the N N F {\displaystyle NNF} . The iterations examine the offsets in scan order (from left to right, top to bottom), and each undergoes propagation followed by random search. === Propagation === We attempt to improve f ( x , y ) {\displaystyle f(x,y)} using the known offsets of f ( x − 1 , y ) {\displaystyle f(x-1,y)} and f ( x , y − 1 ) {\displaystyle f(x,y-1)} , assuming that the patch offsets are likely to be the same. That is, the algorithm will take new value for f ( x , y ) {\displaystyle f(x,y)} to be arg ⁡ min ( x , y ) D ( f ( x , y ) ) , D ( f ( x − 1 , y ) ) , D ( f ( x , y − 1 ) ) {\displaystyle \arg \min \limits _{(x,y)}{D(f(x,y)),D(f(x-1,y)),D(f(x,y-1))}} . So if f ( x , y ) {\displaystyle f(x,y)} has a correct mapping and is in a coherent region R {\displaystyle R} , then all of R {\displaystyle R} below and to the right of f ( x , y ) {\displaystyle f(x,y)} will be filled with the correct mapping. Alternatively, on even iterations, the algorithm search for different direction, fill the new value to be arg ⁡ min ( x , y ) { D ( f ( x , y ) ) , D ( f ( x + 1 , y ) ) , D ( f ( x , y + 1 ) ) } {\displaystyle \arg \min \limits _{(x,y)}\{D(f(x,y)),D(f(x+1,y)),D(f(x,y+1))\}} . === Random search === Let v 0 = f ( x , y ) {\displaystyle v_{0}=f(x,y)} , we attempt to improve f ( x , y ) {\displaystyle f(x,y)} by testing a sequence of candidate offsets at an exponentially decreasing distance from v 0 {\displaystyle v_{0}} u i = v 0 + w α i R i {\displaystyle u_{i}=v_{0}+w\alpha ^{i}R_{i}} where R i {\displaystyle R_{i}} is a uniform random in [ − 1 , 1 ] × [ − 1 , 1 ] {\displaystyle [-1,1]\times [-1,1]} , w {\displaystyle w} is a large window search radius which will be set to maximum picture size, and α {\displaystyle \alpha } is a fixed ratio often assigned as 1/2. This part of the algorithm allows the f ( x , y ) {\displaystyle f(x,y)} to jump out of local minimum through random process. === Halting criterion === The often used halting criterion is set the iteration times to be about 4~5. Even with low iteration, the algorithm works well.

  Read more →
• 

  Dynamic texture

  Dynamic texture ( sometimes referred to as temporal texture) is the texture with motion which can be found in videos of sea-waves, fire, smoke, wavy trees, etc. Dynamic texture has a spatially repetitive pattern with time-varying visual pattern. Modeling and analyzing dynamic texture is a topic of images processing and pattern recognition in computer vision. Extracting features that describe the dynamic texture can be utilized for tasks of images sequences classification, segmentation, recognition and retrieval. Comparing with texture found within static images, analyzing dynamic texture is a challenging problem. It is important that the extracted features from dynamic texture combine motion and appearance description, and also be invariance to some transformation such as rotation, translation and illumination. == Analysis methods of dynamic texture == The methods of dynamic texture recognition can categorized as follows: Methods based on optical flow: by applying optical flow to the dynamic texture, velocity with direction and magnitude can be detected and used to recognize the dynamic texture. Due to simplicity of its computation, it is currently the most popular method. Methods computing geometric properties: this methods track the surfaces of motion trajectories in spatiotemporal domain. Methods based on local spatiotemporal filtering : this methods analyze the local spatiotemporal patterns and its orientation and energy and employ them as feature used for classification. Methods based on global spatiotemporal transform: this method characterize the motion at different scale using wavelets that can decompose the motion into local and global. Model-based methods : These methods aims at generating a model to describe the motion by a set of parameters. == Applications == - Segmenting the sequence images of natural scenes. This helps on differentiate between streets and grass alongside these streets which could be used in the application of navigations. - Motion detection : Dynamic texture features extracted from footage videos can be exploited to detect abnormal crowd activities. - Video classification: video of natural scenes or other scenes that exhibit dynamic textures. - Video retrieval : Dynamic textures can be employed as a feature retrieve videos that contain, for example, sea-waves, smoke, clouds, wavy trees.

  Read more →
• 

  PeduliLindungi

  SatuSehat (Indonesian for "one health"), formerly PeduliLindungi (roughly "care to protect"), is a national integrated health data exchange platform, jointly developed by the Indonesian Ministry of Communication and Information Technology (Kemenkominfo), in partnership with Committee for COVID-19 Response and National Economic Recovery (KPCPEN), Ministry of Health (Kemenkes), Ministry of State-Owned Enterprises (KemenBUMN), and Telkom Indonesia. The SatuSehat platform aims to facilitate data accessibility and service efficiency for health providers and the government, and assist the public as a tool to access their own electronic medical record data. This app was the official COVID-19 contact tracing app used for digital contact tracing in Indonesia, and originally known as TraceTogether but later changed because Singapore had its app using the same name. == Implementation == On 23 August 2021, Coordinating Minister for Maritime and Investments Affairs, Luhut Binsar Panjaitan, encouraged the government to make this app a mandatory requirement before using public transportations, such as train, bus, ferry, and plane. Furthermore, citizen must have installed the app before entering shopping malls, factories, and sport venues. Every person who have received at least a dose of vaccine will receive a vaccine card and vaccination certificate which can be downloaded from the app. In December 2022, with the revocation of PPKM (Community Activities Restrictions Enforcement) starting from 1 January 2023, Ministry of Health issued a statement that the usage of the app is not a governmental mandatory requirement as it used to be. === Transition into a citizen health app === On 7 September 2022, it was announced that the app would be modified to become a citizen health app, capitalising on the reach of the app and the existing work done around the app. On 28 February 2023, the authorities announced that the app was rebranded to SATUSEHAT Mobile (lit. 'OneHealth Mobile'), with existing users needing to update the PeduliLindungi app and re-synchronise their COVID-19 related health information. The re-branded app would eventually be an all-in-one health service and records retrieval app for Indonesians. == Controversy == It was reported that the app requires continuous access to the phone's files, media, and GPS, which quickly drains the battery. Allowing location access only during use or denying it altogether will render the app unusable. This stands in stark contrast to COVID-19 apps used in other countries that only utilize Bluetooth and do not require any additional permissions. In September 2021, stored personal data of at least 1.3 million Indonesian residents were leaked online, including the vaccine certificate of President Joko Widodo. The data leak was also reported on eHAC (electronic Health Alert Card), a mandatory app used for air passengers.

  Read more →
• 

  Shape context

  Shape context is a feature descriptor used in object recognition. Serge Belongie and Jitendra Malik proposed the term in their paper "Matching with Shape Contexts" in 2000. == Theory == The shape context is intended to be a way of describing shapes that allows for measuring shape similarity and the recovering of point correspondences. The basic idea is to pick n points on the contours of a shape. For each point pi on the shape, consider the n − 1 vectors obtained by connecting pi to all other points. The set of all these vectors is a rich description of the shape localized at that point but is far too detailed. The key idea is that the distribution over relative positions is a robust, compact, and highly discriminative descriptor. So, for the point pi, the coarse histogram of the relative coordinates of the remaining n − 1 points, h i ( k ) = # { q ≠ p i : ( q − p i ) ∈ bin ( k ) } {\displaystyle h_{i}(k)=\#\{q\neq p_{i}:(q-p_{i})\in {\mbox{bin}}(k)\}} is defined to be the shape context of p i {\displaystyle p_{i}} . The bins are normally taken to be uniform in log-polar space. The fact that the shape context is a rich and discriminative descriptor can be seen in the figure below, in which the shape contexts of two different versions of the letter "A" are shown. (a) and (b) are the sampled edge points of the two shapes. (c) is the diagram of the log-polar bins used to compute the shape context. (d) is the shape context for the point marked with a circle in (a), (e) is that for the point marked as a diamond in (b), and (f) is that for the triangle. As can be seen, since (d) and (e) are the shape contexts for two closely related points, they are quite similar, while the shape context in (f) is very different. For a feature descriptor to be useful, it needs to have certain invariances. In particular it needs to be invariant to translation, scaling, small perturbations, and, depending on the application, rotation. Translational invariance comes naturally to shape context. Scale invariance is obtained by normalizing all radial distances by the mean distance α {\displaystyle \alpha } between all the point pairs in the shape although the median distance can also be used. Shape contexts are empirically demonstrated to be robust to deformations, noise, and outliers using synthetic point set matching experiments. One can provide complete rotational invariance in shape contexts. One way is to measure angles at each point relative to the direction of the tangent at that point (since the points are chosen on edges). This results in a completely rotationally invariant descriptor. But of course this is not always desired since some local features lose their discriminative power if not measured relative to the same frame. Many applications in fact forbid rotational invariance e.g. distinguishing a "6" from a "9". == Use in shape matching == A complete system that uses shape contexts for shape matching consists of the following steps (which will be covered in more detail in the Details of Implementation section): Randomly select a set of points that lie on the edges of a known shape and another set of points on an unknown shape. Compute the shape context of each point found in step 1. Match each point from the known shape to a point on an unknown shape. To minimize the cost of matching, first choose a transformation (e.g. affine, thin plate spline, etc.) that warps the edges of the known shape to the unknown (essentially aligning the two shapes). Then select the point on the unknown shape that most closely corresponds to each warped point on the known shape. Calculate the "shape distance" between each pair of points on the two shapes. Use a weighted sum of the shape context distance, the image appearance distance, and the bending energy (a measure of how much transformation is required to bring the two shapes into alignment). To identify the unknown shape, use a nearest-neighbor classifier to compare its shape distance to shape distances of known objects. == Details of implementation == === Step 1: Finding a list of points on shape edges === The approach assumes that the shape of an object is essentially captured by a finite subset of the points on the internal or external contours on the object. These can be simply obtained using the Canny edge detector and picking a random set of points from the edges. Note that these points need not and in general do not correspond to key-points such as maxima of curvature or inflection points. It is preferable to sample the shape with roughly uniform spacing, though it is not critical. === Step 2: Computing the shape context === This step is described in detail in the Theory section. === Step 3: Computing the cost matrix === Consider two points p and q that have normalized K-bin histograms (i.e. shape contexts) g(k) and h(k). As shape contexts are distributions represented as histograms, it is natural to use the χ2 test statistic as the "shape context cost" of matching the two points: C S = 1 2 ∑ k = 1 K [ g ( k ) − h ( k ) ] 2 g ( k ) + h ( k ) {\displaystyle C_{S}={\frac {1}{2}}\sum _{k=1}^{K}{\frac {[g(k)-h(k)]^{2}}{g(k)+h(k)}}} The values of this range from 0 to 1. In addition to the shape context cost, an extra cost based on the appearance can be added. For instance, it could be a measure of tangent angle dissimilarity (particularly useful in digit recognition): C A = 1 2 ‖ ( cos ⁡ ( θ 1 ) sin ⁡ ( θ 1 ) ) − ( cos ⁡ ( θ 2 ) sin ⁡ ( θ 2 ) ) ‖ {\displaystyle C_{A}={\frac {1}{2}}{\begin{Vmatrix}{\dbinom {\cos(\theta _{1})}{\sin(\theta _{1})}}-{\dbinom {\cos(\theta _{2})}{\sin(\theta _{2})}}\end{Vmatrix}}} This is half the length of the chord in unit circle between the unit vectors with angles θ 1 {\displaystyle \theta _{1}} and θ 2 {\displaystyle \theta _{2}} . Its values also range from 0 to 1. Now the total cost of matching the two points could be a weighted-sum of the two costs: C = ( 1 − β ) C S + β C A {\displaystyle C=(1-\beta )C_{S}+\beta C_{A}\!\,} Now for each point pi on the first shape and a point qj on the second shape, calculate the cost as described and call it Ci,j. This is the cost matrix. === Step 4: Finding the matching that minimizes total cost === Now, a one-to-one matching π ( i ) {\displaystyle \pi (i)} that matches each point pi on shape 1 and qj on shape 2 that minimizes the total cost of matching, H ( π ) = ∑ i C ( p i , q π ( i ) ) {\displaystyle H(\pi )=\sum _{i}C\left(p_{i},q_{\pi (i)}\right)} is needed. This can be done in O ( N 3 ) {\displaystyle O(N^{3})} time using the Hungarian method, although there are more efficient algorithms. To have robust handling of outliers, one can add "dummy" nodes that have a constant but reasonably large cost of matching to the cost matrix. This would cause the matching algorithm to match outliers to a "dummy" if there is no real match. === Step 5: Modeling transformation === Given the set of correspondences between a finite set of points on the two shapes, a transformation T : R 2 → R 2 {\displaystyle T:\mathbb {R} ^{2}\to \mathbb {R} ^{2}} can be estimated to map any point from one shape to the other. There are several choices for this transformation, described below. ==== Affine ==== The affine model is a standard choice: T ( p ) = A p + o {\displaystyle T(p)=Ap+o\!} . The least squares solution for the matrix A {\displaystyle A} and the translational offset vector o is obtained by: o = 1 n ∑ i = 1 n ( p i − q π ( i ) ) , A = ( Q + P ) t {\displaystyle o={\frac {1}{n}}\sum _{i=1}^{n}\left(p_{i}-q_{\pi (i)}\right),A=(Q^{+}P)^{t}} Where P = ( 1 p 11 p 12 ⋮ ⋮ ⋮ 1 p n 1 p n 2 ) {\displaystyle P={\begin{pmatrix}1&p_{11}&p_{12}\\\vdots &\vdots &\vdots \\1&p_{n1}&p_{n2}\end{pmatrix}}} with a similar expression for Q {\displaystyle Q\!} . Q + {\displaystyle Q^{+}\!} is the pseudoinverse of Q {\displaystyle Q\!} . ==== Thin plate spline ==== The thin plate spline (TPS) model is the most widely used model for transformations when working with shape contexts. A 2D transformation can be separated into two TPS function to model a coordinate transform: T ( x , y ) = ( f x ( x , y ) , f y ( x , y ) ) {\displaystyle T(x,y)=\left(f_{x}(x,y),f_{y}(x,y)\right)} where each of the ƒx and ƒy have the form: f ( x , y ) = a 1 + a x x + a y y + ∑ i = 1 n ω i U ( ‖ ( x i , y i ) − ( x , y ) ‖ ) , {\displaystyle f(x,y)=a_{1}+a_{x}x+a_{y}y+\sum _{i=1}^{n}\omega _{i}U\left({\begin{Vmatrix}(x_{i},y_{i})-(x,y)\end{Vmatrix}}\right),} and the kernel function U ( r ) {\displaystyle U(r)\!} is defined by U ( r ) = r 2 log ⁡ r 2 {\displaystyle U(r)=r^{2}\log r^{2}\!} . The exact details of how to solve for the parameters can be found elsewhere but it essentially involves solving a linear system of equations. The bending energy (a measure of how much transformation is needed to align the points) will also be easily obtained. ==== Regularized TPS ==== The TPS formulation above has exact matching requirement for the pairs of points on the two shapes. For noisy data, it is best to

  Read more →
• 

  Local ternary patterns

  Local ternary patterns (LTP) are an extension of local binary patterns (LBP). Unlike LBP, it does not threshold the pixels into 0 and 1, rather it uses a threshold constant to threshold pixels into three values. Considering k as the threshold constant, c as the value of the center pixel, a neighboring pixel p, the result of threshold is: { 1 , if p > c + k 0 , if p > c − k and p < c + k − 1 if p < c − k {\displaystyle {\begin{cases}1,&{\text{if }}p>c+k\\0,&{\text{if }}p>c-k{\text{ and }}p Read more →

• 

  Neural processing unit

  A neural processing unit (NPU), also known as an AI accelerator or deep learning processor, is a class of specialized hardware accelerator or computer system designed to accelerate artificial intelligence and machine learning applications, including artificial neural networks and computer vision. == Use == Their purpose is either to efficiently execute already trained AI models (inference) or to train AI models. NPUs can be more efficient in terms of speed or power consumption. NPU applications include algorithms for robotics, Internet of things, and data-intensive or sensor-driven tasks. They are often manycore or spatial designs and focus on low-precision arithmetic, novel dataflow architectures, or in-memory computing capability. As of 2024, a widely used datacenter-grade AI integrated circuit chip, the Nvidia H100 GPU, contains tens of billions of MOSFETs. === Consumer devices === AI accelerators are used in Apple silicon, Qualcomm, Samsung, Huawei, and Google Tensor smartphone processors. Vision processing units are accelerators specialized for machine vision algorithms such as CNN (convolutional neural networks) and SIFT (scale-invariant feature transform). They are used in devices that need to keep track of objects visually such as AR headsets and drones. It is more recently (circa 2017) added to processors from Apple and (circa 2022) to processors from Intel and AMD. All models of Intel Meteor Lake processors have a built-in versatile processor unit (VPU) for accelerating inference for computer vision and deep learning. On consumer devices, the NPU is intended to be small, power-efficient, but reasonably fast when used to run small models. To do this they are designed to support low-bitwidth operations using data types such as INT4, INT8, FP8, and FP16. A common metric is trillions of operations per second (TOPS). Although TOPS does not explicitly specify the kind of operations, it is typically INT8 additions and multiplications. === Datacenters === Accelerators are used in cloud computing servers: e.g., tensor processing units (TPU) for Google Cloud Platform, and Trainium and Inferentia chips for Amazon Web Services. Many vendor-specific terms exist for devices in this category, and it is an emerging technology without a dominant design. Since the late 2010s, graphics processing units designed by companies such as Nvidia and AMD often include AI-specific hardware in the form of dedicated functional units for low-precision matrix-multiplication operations. These GPUs are commonly used as AI accelerators, both for training and inference. === Scientific computation === Although NPUs are tailored for low-precision (e.g., FP16, INT8) matrix multiplication operations, they can be used to emulate higher-precision matrix multiplications in scientific computing. As modern GPUs place much focus on making the NPU part fast, using emulated FP64 (Ozaki scheme) on NPUs can potentially outperform native FP64. This has been demonstrated using FP16-emulated FP64 on NVIDIA TITAN RTX and using INT8-emulated FP64 on NVIDIA consumer GPUs and the A100 GPU. Consumer GPUs especially benefited as they have limited FP64 hardware capacity, showing a 6× speedup. Since CUDA Toolkit 13.0 Update 2, cuBLAS automatically uses INT8-emulated FP64 matrix multiplication of the equivalent precision if it is faster than native. This is in addition to the FP16-emulated FP32 feature introduced in version 12.9. == Programming == An operating system or a higher-level library may provide application programming interfaces such as TensorFlow with LiteRT Next (Android), CoreML (iOS, macOS) or DirectML (Windows). Formats such as ONNX are used to represent trained neural networks. Consumer CPU-integrated NPUs are accessible through vendor-specific APIs. AMD (Ryzen AI), Intel (OpenVINO), Apple silicon (CoreML), and Qualcomm (SNPE) each have their own APIs, which can be built upon by a higher-level library. GPUs generally use existing GPGPU pipelines such as CUDA and OpenCL adapted for lower precisions and specialized matrix-multiplication operations. Vulkan is also being used. Custom-built systems such as the Google TPU use private interfaces. There are a large number of separate underlying acceleration APIs and compilers/runtimes in use in the AI field, causing a great increase in software development effort due to the many combinations involved. As of 2025, the open standard organization Khronos Group is pursuing standardization of AI-related interfaces to reduce the amount of work needed. Khronos is working on three separate fronts: expansion of data types and intrinsic operations in OpenCL and Vulkan, inclusion of compute graphs in SPIR-V, and a NNEF/SkriptND file format for describing a neural network.

  Read more →
• 

  Tinybop

  Tinybop is a Brooklyn based publisher of apps for children. == History == Tinybop is a Brooklyn-based children's media company established in 2011 by Raul Gutierrez. App titles are released in two series: the Explorer's Library - a series of science apps and Digital Toys - series of open-ended construction apps. == Published apps == Explorer's Library Titles: The Human Body – An anatomy app for children. Released 2013. The company's first app was illustrated by Kelli Anderson and has been downloaded millions of times. Selected for the American Library Association's Notable Children's Media List in 2022. Named Apple App Store's Best of 2013. Winner of the Digital Ehon Yuichi Kimura Prize for Children's Digital Media. Plants – An app about biomes around the world. Homes – An app about houses around with world. Illustrated by Tuesday Bassen. Winner of the Parents Gold Choice Award for children's apps. Simple Machines – A children's physics app about simple machines. The Earth – An app for children about the geologic Earth illustrated by Sarah Jacoby. Weather – A children's weather app. Skyscrapers – A children's app about building tall buildings. Space – An interactive solar system. Mammals – A children's app about mammals illustrated by Wenjia Tang. Winner of the Digital Ehon Award for Children's Educational media. Coral Reef – An app about marine ecosystems. Winner of an Excellence in Early Learning Digital Media Honor from the American Library Association. State of Matter – An app covering solids, liquids, and gases. Winner of Excellence in Early Learning Digital Media Honor from the American Library Association. Light and Color – An app about light and color. Selected for The American Library Association's Notable Children's Media List 2023. Winner of the 2022 Yoichi Sakakihara Prize for Children's Media. Digital Toys Titles: The Robot Factory – A robot building app for children illustrated by Owen Davey. Apple named The Robot Factory as iPad App of the Year in 2015. The Everything Machine – A visual coding app for children. The Everything Machine was named Apple's Best of 2015. Monsters – A monster creation app illustrated by Tianhua Mao. The Infinite Arcade – An arcade game building app. Me: A Kids Diary – A digital journal for children. Selected for The American Library Association's Notable Children's Media List 2020. The Creature Garden – An app that allows children to create fantastical animals illustrated by Natasha Durley. Selected for The American Library Association's Notable Children's Media List 2021. Things that Go Bump – A multiplayer game set in an enchanted Japanese house, released on Apple Arcade in 2018.

  Read more →
• 

  ChatScript

  ChatScript is a combination Natural Language engine and dialog management system designed initially for creating chatbots, but is currently also used for various forms of NL processing. It is written in C++. The engine is an open source project at SourceForge. and GitHub. ChatScript was written by Bruce Wilcox and originally released in 2011, after Suzette (written in ChatScript) won the 2010 Loebner Prize, fooling one of four human judges. == Features == In general ChatScript aims to author extremely concisely, since the limiting scalability of hand-authored chatbots is how much/fast one can write the script. Because ChatScript is designed for interactive conversation, it automatically maintains user state across volleys. A volley is any number of sentences the user inputs at once and the chatbots response. The basic element of scripting is the rule. A rule consists of a type, a label (optional), a pattern, and an output. There are three types of rules. Gambits are something a chatbot might say when it has control of the conversation. Rejoinders are rules that respond to a user remark tied to what the chatbot just said. Responders are rules that respond to arbitrary user input which is not necessarily tied to what the chatbot just said. Patterns describe conditions under which a rule may fire. Patterns range from extremely simplistic to deeply complex (analogous to Regex but aimed for NL). Heavy use is typically made of concept sets, which are lists of words sharing a meaning. ChatScript contains some 2000 predefined concepts and scripters can easily write their own. Output of a rule intermixes literal words to be sent to the user along with common C-style programming code. Rules are bundled into collections called topics. Topics can have keywords, which allows the engine to automatically search the topic for relevant rules based on user input. == Example code == Words starting with ~ are concept sets. For example, ~fruit is the list of all known fruits. The simple pattern (~fruit) reacts if any fruit is mentioned immediately after the chatbot asks for favorite food. The slightly more complex pattern for the rule labelled WHATMUSIC requires all the words what, music, you and any word or phrase meaning to like, but they may occur in any order. Responders come in three types. ?: rules react to user questions. s: rules react to user statements. u: rules react to either. ChatScript code supports standard if-else, loops, user-defined functions and calls, and variable assignment and access. == Data == Some data in ChatScript is transient, meaning it will disappear at the end of the current volley. Other data is permanent, lasting forever until explicitly killed off. Data can be local to a single user or shared across all users at the bot level. Internally all data is represented as text and is automatically converted to a numeric form as needed. === Variables === User variables come in several kinds. Variables purely local to a topic or function are transient. Global variables can be declared as transient or permanent. A variable is generally declared merely by using it, and its type depends on its prefix ($, $$, $_). === Facts === In addition to variables, ChatScript supports facts – triples of data, which can also be transient or permanent. Functions can query for facts having particular values of some of the fields, making them act like an in-memory database. Fact retrieval is very quick and efficient the number of available in-memory facts is largely constrained to the available memory of the machine running the ChatScript engine. Facts can represent record structures and are how ChatScript represents JSON internally. Tables of information can be defined to generate appropriate facts. The above table links people to what they invented (1 per line) with Einstein getting a list of things he did. == External communication == ChatScript embeds the Curl library and can directly read and write facts in JSON to a website. == Server == A ChatScript engine can run in local or server mode. == Pos-tagging, parsing, and ontology == ChatScript comes with a copy of English WordNet embedded within, including its ontology, and creates and extends its own ontology via concept declarations. It has an English language pos-tagger and parser and supports integration with TreeTagger for pos-tagging a number of other languages (TreeTagger commercial license required). == Databases == In addition to an internal fact database, ChatScript supports PostgreSQL, MySQL, MSSQL and MongoDB both for access by scripts, but also as a central filesystem if desired so ChatScript can be scaled horizontally. A common use case is to use a centralized database to host the user files and multiple servers to scale the ChatScript engine. == JavaScript == ChatScript also embeds DukTape, ECMAScript E5/E5.1 compatibility, with some semantics updated from ES2015+. == Spelling Correction == ChatScript has built-in automatic spell checking, which can be augmented in script as both simple word replacements or context sensitive changes. With appropriate simple rules you can change perfect legal words into other words or delete them. E.g., if you have a concept of ~electronic_goods and don't want an input of Radio Shack (a store name) to be detected as an electronic good, you can get the input to change to Radio_Shack (a single word), or allow the words to remain but block the detection of the concept. This is particularly useful when combined with speech-to-text code that is imperfect, but you are familiar with common failings of it and can compensate for them in script. == Control flow == A chatbot's control flow is managed by the control script. This is merely another ordinary topic of rules, that invokes API functions of the engine. Thus control is fully configurable by the scripter (and functions exist to allow introspection into the engine). There are pre-processing control flow and post-processing control flow options available, for special processing.

  Read more →
• 

  Visual temporal attention

  Visual temporal attention is a special case of visual attention that involves directing attention to specific instant of time. Similar to its spatial counterpart visual spatial attention, these attention modules have been widely implemented in video analytics in computer vision to provide enhanced performance and human interpretable explanation of deep learning models. As visual spatial attention mechanism allows human and/or computer vision systems to focus more on semantically more substantial regions in space, visual temporal attention modules enable machine learning algorithms to emphasize more on critical video frames in video analytics tasks, such as human action recognition. In convolutional neural network-based systems, the prioritization introduced by the attention mechanism is regularly implemented as a linear weighting layer with parameters determined by labeled training data. == Application in Action Recognition == Recent video segmentation algorithms often exploits both spatial and temporal attention mechanisms. Research in human action recognition has accelerated significantly since the introduction of powerful tools such as Convolutional Neural Networks (CNNs). However, effective methods for incorporation of temporal information into CNNs are still being actively explored. Motivated by the popular recurrent attention models in natural language processing, the Attention-aware Temporal Weighted CNN (ATW CNN) is proposed in videos, which embeds a visual attention model into a temporal weighted multi-stream CNN. This attention model is implemented as temporal weighting and it effectively boosts the recognition performance of video representations. Besides, each stream in the proposed ATW CNN framework is capable of end-to-end training, with both network parameters and temporal weights optimized by stochastic gradient descent (SGD) with back-propagation. Experimental results show that the ATW CNN attention mechanism contributes substantially to the performance gains with the more discriminative snippets by focusing on more relevant video segments. == Literature == Seibold VC, Balke J and Rolke B (2023): Temporal attention. Front. Cognit. 2:1168320. doi: 10.3389/fcogn.2023.1168320.

  Read more →
• 

  IruSoft

  IruSoft (Arabic: آيروسوفت) is an insurance regulatory platform designated for licensing, supervision and inspection of the insurance sector within a country. The platform introduced unique supervision-technology (suptech), insurance-technology (insurtech) and regulatory-technology (regtech) automated modules by which a regulator requires less resources to ensure fairness, transparency and competition and to prevent conflicts of interest in the sector. IruSoft was founded by Abdullah Al-Salloum and owned by the Insurance Regulatory Unit in Kuwait. The Insurance Regulatory Unit optimized processing insurance-sector's customer complaints by issuing Resolution No. (1) of 2022 that introduced IruSoft's complaints public module; an automated resolution center, by which the process of receiving submitted complaints, passing them on to the platforms of licensed insurance companies, tracking matter-related discussions and updates and getting them escalated if unresolved to be discussed by a committee assigned by the unit is integrally automated and analyzed for better key performance indicators.

  Read more →
• 

  Cloud robotics

  Cloud robotics is a field of robotics that attempts to invoke cloud technologies such as cloud computing, cloud storage, and other Internet technologies centered on the benefits of converged infrastructure and shared services for robotics. When connected to the cloud, robots can benefit from the powerful computation, storage, and communication resources of a modern data center in the cloud, which can process and share information from various robots or agents (other machines, smart objects, humans, etc.). Humans can also delegate tasks to robots remotely through networks. Cloud computing technologies enable robot systems to be gain capability whilst reducing costs through cloud technologies. Thus, it is possible to build lightweight, low-cost, smarter robots with an intelligent "brain" in the cloud. The "brain" consists of data center, knowledge base, task planners, deep learning, information processing, environment models, communication support, etc. == Components == A cloud for robots potentially has at least six significant components: Building a "cloud brain" for robots, the main object of cloud robotics; Offering a global library of images, maps, and object data, often with geometry and mechanical properties, expert system, knowledge base (i.e. semantic web, data centres); Massively-parallel computation on demand for sample-based statistical modelling and motion planning, task planning, multi-robot collaboration, scheduling and coordination of system; Robot sharing of outcomes, trajectories, and dynamic control policies and robot learning support; Human sharing of open-source code, data, and designs for programming, experimentation, and hardware construction; On-demand human guidance and assistance for evaluation, learning, and error recovery; Augmented human–robot interaction through various ways (semantics knowledge base, Apple SIRI like service, etc.). == Applications == Autonomous mobile robots Google's self-driving cars are cloud robots. The cars use the network to access Google's enormous database of maps and satellite and environment model (like Streetview) and combines it with streaming data from GPS, cameras, and 3D sensors to monitor its own position within centimetres, and with past and current traffic patterns to avoid collisions. Each car can learn something about environments, roads, or driving, or conditions, and it sends the information to the Google cloud, where it can be used to improve the performance of other cars. Cloud medical robots a medical cloud (also called a healthcare cluster) consists of various services such as a disease archive, electronic medical records, a patient health management system, practice services, analytics services, clinic solutions, expert systems, etc. A robot can connect to the cloud to provide clinical service to patients, as well as deliver assistance to doctors (e.g. a co-surgery robot). Moreover, it also provides a collaboration service by sharing information between doctors and care givers about clinical treatment. Assistive robots A domestic robot can be employed for healthcare and life monitoring for elderly people. The system collects the health status of users and exchange information with cloud expert system or doctors to facilitate elderly peoples life, especially for those with chronic diseases. For example, the robots are able to provide support to prevent the elderly from falling down, emergency healthy support such as heart disease, blooding disease. Care givers of elderly people can also get notification when in emergency from the robot through network. Industrial robots As highlighted by the German government's Industry 4.0 Plan, "Industry is on the threshold of the fourth industrial revolution. Driven by the Internet, the real and virtual worlds are growing closer and closer together to form the Internet of Things. Industrial production of the future will be characterised by the strong individualisation of products under the conditions of highly flexible (large series) production, the extensive integration of customers and business partners in business and value-added processes, and the linking of production and high-quality services leading to so-called hybrid products." In manufacturing, such cloud based robot systems could learn to handle tasks such as threading wires or cables, or aligning gaskets from a professional knowledge base. A group of robots can share information for some collaborative tasks. Even more, a consumer is able to place customised product orders to manufacturing robots directly with online ordering systems. Another potential paradigm is shopping-delivery robot systems. Once an order is placed, a warehouse robot dispatches the item to an autonomous car or autonomous drone to deliver it to its recipient. == Research == RoboEarth was funded by the European Union's Seventh Framework Programme for research, technological development projects, specifically to explore the field of cloud robotics. The goal of RoboEarth is to allow robotic systems to benefit from the experience of other robots, paving the way for rapid advances in machine cognition and behaviour, and ultimately, for more subtle and sophisticated human-machine interaction. RoboEarth offers a Cloud Robotics infrastructure. RoboEarth's World-Wide-Web style database stores knowledge generated by humans – and robots – in a machine-readable format. Data stored in the RoboEarth knowledge base include software components, maps for navigation (e.g., object locations, world models), task knowledge (e.g., action recipes, manipulation strategies), and object recognition models (e.g., images, object models). The RoboEarth Cloud Engine includes support for mobile robots, autonomous vehicles, and drones, which require much computation for navigation. Rapyuta is an open source cloud robotics framework based on RoboEarth Engine developed by the robotics researcher at ETHZ. Within the framework, each robot connected to Rapyuta can have a secured computing environment (rectangular boxes) giving them the ability to move their heavy computation into the cloud. In addition, the computing environments are tightly interconnected with each other and have a high bandwidth connection to the RoboEarth knowledge repository. FogROS2 is an open-source extension to the Robot Operating System 2 (ROS 2) developed by researchers at UC Berkeley. It enables robots to offload computationally intensive tasks—such as SLAM, grasp planning, and motion planning—to cloud resources, thereby enhancing performance and reducing onboard computational requirements. FogROS2 automates the provisioning of cloud instances, deployment of ROS 2 nodes, and secure communication between robots and cloud services. The platform is designed to be compatible with existing ROS 2 applications without requiring code modifications. Further advancements include FogROS2-SGC, which facilitates secure global connectivity across different networks and locations, and FogROS2-FT, which introduces fault tolerance by replicating services across multiple cloud providers to ensure robustness against failures. KnowRob is an extensional project of RoboEarth. It is a knowledge processing system that combines knowledge representation and reasoning methods with techniques for acquiring knowledge and for grounding the knowledge in a physical system and can serve as a common semantic framework for integrating information from different sources. RoboBrain is a large-scale computational system that learns from publicly available Internet resources, computer simulations, and real-life robot trials. It accumulates everything robotics into a comprehensive and interconnected knowledge base. Applications include prototyping for robotics research, household robots, and self-driving cars. The goal is as direct as the project's name—to create a centralised, always-online brain for robots to tap into. The project is dominated by Stanford University and Cornell University. And the project is supported by the National Science Foundation, the Office of Naval Research, the Army Research Office, Google, Microsoft, Qualcomm, the Alfred P. Sloan Foundation and the National Robotics Initiative, whose goal is to advance robotics to help make the United States more competitive in the world economy. MyRobots is a service for connecting robots and intelligent devices to the Internet. It can be regarded as a social network for robots and smart objects (i.e. Facebook for robots). With socialising, collaborating and sharing, robots can benefit from those interactions too by sharing their sensor information giving insight on their perspective of their current state. COALAS is funded by the INTERREG IVA France (Channel) – England European cross-border co-operation programme. The project aims to develop new technologies for disabled people through social and technological innovation and through the users' social and psychological integrity. The objective is to produce a cognitive ambient

  Read more →
• 

  1.58-bit large language model

  A 1.58-bit large language model (also known as a ternary LLM) is a type of large language model (LLM) designed to be computationally efficient. It achieves this by using weights that are restricted to only three values: -1, 0, and +1. This restriction significantly reduces the model's memory footprint and allows for faster processing, as computationally expensive multiplication operations can be replaced with lower-cost additions. This contrasts with traditional models that use 16-bit floating-point numbers (FP16 or BF16) for their weights. Studies have shown that for models up to several billion parameters, the performance of 1.58-bit LLMs on various tasks is comparable to their full-precision counterparts. This approach could enable powerful AI to run on less specialized and lower-power hardware. The name "1.58-bit" comes from the fact that a system with three states contains log 2 ⁡ 3 ≈ 1.58 {\displaystyle \log _{2}3\approx 1.58} bits of information. These models are sometimes also referred to as 1-bit LLMs in research papers, although this term can also refer to true binary models (with weights of -1 and +1). == BitNet == In 2024, Ma et al., researchers at Microsoft, declared that their 1.58-bit model, BitNet b1.58 is comparable in performance to the 16-bit Llama 2 and opens the era of 1-bit LLM. BitNet creators did not use the post-training quantization of weights but instead relied on the new BitLinear transform that replaced the nn.Linear layer of the traditional transformer design. In 2025, Microsoft researchers had released an open-weights and open inference code model BitNet b1.58 2B4T demonstrating performance competitive with the full precision models at 2B parameters and 4T training tokens. == Post-training quantization == BitNet derives its performance from being trained natively in 1.58 bit instead of being quantized from a full-precision model after training. Still, training is an expensive process and it would be desirable to be able to somehow convert an existing model to 1.58 bits. In 2024, HuggingFace reported a way to gradually ramp up the 1.58-bit quantization in fine-tuning an existing model down to 1.58 bits. == Critique == Some researchers point out that the scaling laws of large language models favor the low-bit weights only in case of undertrained models. As the number of training tokens increases, the deficiencies of low-bit quantization surface.

  Read more →
• 

  N-jet

  An N-jet is the set of (partial) derivatives of a function f ( x ) {\displaystyle f(x)} up to order N. Specifically, in the area of computer vision, the N-jet is usually computed from a scale space representation L {\displaystyle L} of the input image f ( x , y ) {\displaystyle f(x,y)} , and the partial derivatives of L {\displaystyle L} are used as a basis for expressing various types of visual modules. For example, algorithms for tasks such as feature detection, feature classification, stereo matching, tracking and object recognition can be expressed in terms of N-jets computed at one or several scales in scale space.

  Read more →