{
  "fetched_at": "2026-10-20T01:48:20Z",
  "source_url": "https://www.kaggle.com/competitions/can-i-make-a-wish-detecting-shooting-stars/data",
  "pages": [
    {
      "name": "data-description",
      "content": "<p>This page shows up underneath the data files. </p>\r\n<h2>File descriptions</h2>\r\n<ul>\r\n<li><strong>train.csv</strong> - the training set</li>\r\n<li><strong>test.csv</strong> - the test set</li>\r\n<li><strong>submission.csv</strong> - a sample submission file in the correct format</li>\r\n<li><strong>DatasetPerFeature.zip -</strong> contains the training and testing sets separated per feature type</li>\r\n</ul>\r\n<h2>Data fields</h2>\r\n<ul>\r\n<li><strong>Id</strong> - an anonymous id unique to each image</li>\r\n<li><strong>X1 to X3451</strong> - extracted features, as described previously</li>\r\n<li><strong>Class</strong> - present in the training set, labels the examples</li>\r\n</ul>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "<p><strong>Attention:</strong> <em>according to the competition rules, the winning teams on private data, that will be invited to present their solutions at the conference, were determined in September, 18th. They are (in alphabetic order):</em></p>\r\n<p><em>Artificial Psycho Killers (Humberto &amp; Hugo)</em></p>\r\n<p><em>ML-UFSCAR</em></p>\r\n<p><em>VictorAlmeida, UFF and Petrobras</em></p>\r\n<p>This competition is part of the 1st KDD-BR (Brazilian Knowledge Discovery in Databases competition) event, one of the joint activities of the 2017 editions of the Brazilian Conference on Intelligent Systems (<a href=\"http://www.bracis2017.ufu.br\">BRACIS</a>), the Brazilian Symposium on Databases (<a href=\"http://sbbd.org.br/2017\">SBBD</a>) and the Symposium on Knowledge Discovery, Mining and Learning (<a href=\"http://www.facom.ufu.br/~kdmile\">KDMiLe</a>).</p>\r\n<p>It involves classifying images captured by a monitoring station installed at the <a href=\"http://www.univap.br\">UNIVAP</a> (University of Vale do Paraíba) astronomy observatory, at São José dos Campos, São Paulo, Brazil. This station composes one of the monitoring stations of the <a href=\"http://exoss.org/\">EXOSS</a> Citizen Science organization. This Brazilian non-profit organization aims to monitor meteors (popularly kown as shooting stars) which cross the southern skies. It is a network currently composed of 50 active monitoring stations at various locations of the Brazilian territory. Any interested citizen can apply as a potential contributor.</p>\r\n<p>The data acquisition is made by a simple and low cost video monitoring camera, usually used in security monitoring systems, with a dedicated software. Each time a moving object is detected by the camera, it starts shooting. This includes meteors, but also birds, insects, planes and even lightning thunders. Some days are cloudy, others rainy, there are varying stars in the sky, and there are many other challenging situations. The camera also takes some noisy shootings sometimes. The system records a video and also an image which is a composition of the video frames, which will be used at the competition. Since the interest is to monitor the occurence of meteors, this competition aims to built an automate system for classifying the recorded images into two classes: meteor (example in left image) vs non-meteor (example in right image).</p>\r\n<p><img src=\"https://storage.googleapis.com/kaggle-media/competitions/inclass/6820/media/meteor-ex.jpg\" alt=\"Example of meteor\" width=\"261\" height=\"209\" />     <img src=\"https://storage.googleapis.com/kaggle-media/competitions/inclass/6820/media/meteor-noex.jpg\" alt=\"Example of non-meteor\" width=\"264\" height=\"211\" /></p>\r\n<p>The images were captured in 2017 during the months of April and May and were categorized by Jennifer Nielsen, an Aeronautics Engeenering student at UNIVAP, under the supervision of Dr Irapuan Rodrigues, Physics and Astronomy Professor at UNIVAP. The \"!UFO Capture\" software was used to capture the videos and images and the \"UFO Analyzer\" software was used for cleaning and analyzing the trajectories. Both softwares are part of the <a href=\"http://sonotaco.com/e_index.html\">Sonotaco</a> package.</p>\r\n<p>A set of features have been extracted from regions of interest pointed by the Sonotaco software in the images, as described next.</p>\r\n<h2>Features</h2>\r\n<p>The following features were extracted from the images:</p>\r\n<p>- <strong>AutoColorCorrelogram (features X1 to X768)</strong><br />We define a new image feature called the color correlogram and use it for image indexing and comparison. This feature distills the spatial correlation of colors, and is both effective and inexpensive for content-based image retrieval. The correlogram robustly tolerates large changes in appearance and shape caused by changes in viewing positions, camera zooms, etc. Experimental evidence suggests that this new feature outperforms not only the traditional color histogram method, but also the recently proposed histogram refinement methods for image indexing/retrieval.<br /># More info: https://www.computer.org/csdl/proceedings/cvpr/1997/7822/00/78220762-abs.html</p>\r\n<p>- <strong>CEDD (features X769 to X912) </strong><br />Deals with a new low level feature that is extracted from the images and can be used for indexing and retrieval. This feature is called “Color and Edge Directivity Descriptor” and incorporates color and texture information in a histogram. CEDD size is limited to 54 bytes per image, rendering this descriptor suitable for use in large image databases.<br /># More info: https://link.springer.com/chapter/10.1007/978-3-540-79547-6_30</p>\r\n<p>- <strong>ColorHistogram (features X913 to X976)</strong><br />A RGB/HSB histogram. This color space can be split sensibly in a non-uniform way, to give more detail e.g. to the hues, while not splitting fine on the saturation. In case of using the HSB color space, good results have been reported with 7x2x2 and 7x3x3 splits, for example.<br /># More info: https://en.wikipedia.org/wiki/Color_histogram</p>\r\n<p>- <strong>FCTH (features X977 to X1168)</strong><br />deals with the extraction of a new low level feature that combines, in one histogram, color and texture information. This feature is named FCTH - Fuzzy Color and Texture Histogram - and results from the combination of 3 fuzzy systems. FCTH size is limited to 72 bytes per image, rendering this descriptor suitable for use in large image databases. The proposed feature is appropriate for accurately retrieving images even in distortion cases such as deformations, noise and smoothing.<br /># More info: http://ieeexplore.ieee.org/abstract/document/4556917/?reload=true</p>\r\n<p>- <strong>FuzzyColorHistogram (features X1169 to X1293)</strong><br />To address these general Color Histogram concerns, this version presents a new color histogram representation, called fuzzy color histogram (FCH), <br />by considering the color similarity of each pixel's color associated to all the histogram bins through fuzzy-set membership function.<br /># More info: http://ieeexplore.ieee.org/document/1025168/</p>\r\n<p>- <strong>FuzzyOpponentHistogram (features X1294 to X1869)</strong><br />Simple fuzzy 64 bin Opponent Histogram, based on the Opponent color space as described in van de Sande, Gevers &amp; Snoek (2010) \"Evaluating Color Descriptors for Object and Scene Recognition\".<br /># More info: http://ieeexplore.ieee.org/document/5204091/</p>\r\n<p>- <strong>Gabor (features X1870 to X1929)</strong><br />A Gabor filter, is a linear filter used for texture analysis, which means that it basically analyses whether there are any specific frequency content in the image in specific directions in a localized region around the point or region of analysis. Frequency and orientation representations of Gabor filters are similar to those of the human visual system, and they have been found to be particularly appropriate for texture representation and discrimination.<br /># More info: https://en.wikipedia.org/wiki/Gabor_filter</p>\r\n<p>- <strong>Haralick (features X1930 to X1943)</strong><br />A co-occurrence matrix or co-occurrence distribution is a matrix that is defined over an image to be the distribution of co-occurring pixel values (grayscale values, or colors) at a given offset. Whether considering the intensity or grayscale values of the image or various dimensions of color, the co-occurrence matrix can measure the texture of the image. Because co-occurrence matrices are typically large and sparse, various metrics of the matrix are often taken to get a more useful set of features. Features generated using this technique are called Haralick features.<br /># More info: http://ieeexplore.ieee.org/abstract/document/4309314/</p>\r\n<p>- <strong>Histogram (features X1944 to X2199)</strong><br />A histogram is a graphical representation of the distribution of numerical data. It is an estimate of the probability distribution of a continuous variable (quantitative variable). The first step is to divide the entire range of values into a series of intervals, and then count how many values fall into each interval. Histograms: Gray-Histogram, RGB-Histogram and HSB-Histogram.<br />#More info: https://en.wikipedia.org/wiki/Histogram</p>\r\n<p>- <strong>JCD (features X2200 to X2367)</strong><br />A joint descriptor joining CEDD and FCTH in one histogram. The proposed method utilizes two sets of keywords in order to map the low-level features of the descriptor to the high-level features constituted by these keywords.<br /># More info: https://pdfs.semanticscholar.org/0cb0/72ac8b0787d37d856d4ab1adcf1532c18258.pdf</p>\r\n<p>- <strong>JPEGCoefficientHistogram (features X2368 to X2559)</strong><br />This feature uses the EdgeHistogram descriptor from the MPEG-7 standard.<br />The MPEG-7 Visual Standard under development specifies content-based descriptors that allow users or agents (or search engines) to measure similarity in images or video based on visual criteria, and can be used to efficiently identify, filter, or browse images or video based on visual content.<br />More specifically, MPEG-7 specifies color, texture, object shape, global motion, or object motion features for this purpose.<br /># More info: http://www.ee.columbia.edu/~sfchang/course/vis/REF/sikora01.pdf</p>\r\n<p>- <strong>LuminanceLayout (features X2560 to X2623)</strong><br />The LuminanceLayout Descriptor is intended for grayscale or B/W images. It scales an image down to a very small size and uses this smaller version as a<br />descriptor. Interesting aspect is that white stripes are added to make the small image quadratic.<br /># More info: https://etrij.etri.re.kr/etrij/journal/article/article.do?volume=29&amp;issue=&amp;page=246</p>\r\n<p>- <strong>MeanIntensityLocalBinaryPatterns (features X2624 to X2879)</strong><br />Constructs histogram of binary patterns based on comparing the mean intensity of the 8-neighborhood of each pixel N&lt;sub&gt;i&lt;/sub&gt; in the patch to the mean intensity of the 8-neighborhood around the pixel in the center N&lt;sub&gt;C&lt;/sub&gt;. Each of the 8 mean intensities N&lt;sub&gt;i&lt;/sub&gt; is compared to the mean intensity N&lt;sub&gt;C&lt;/sub&gt; in the center, if N&lt;sub&gt;i&lt;/sub&gt; is smaller than N&lt;sub&gt;C&lt;/sub&gt;, N&lt;sub&gt;i&lt;/sub&gt; is assigned weight 0, else 1. The weights form a binary pattern by contactinating them in counter clockwise order, starting with the weight to the left of the pixel in the center.<br /># More info: http://ieeexplore.ieee.org/document/576366/</p>\r\n<p>- <strong>MeanPatchIntensityHistogram (features X2880 to X3135)</strong><br />This descriptor calculates a histogram of mean intensities of specified neighborhood size. For each pixel in the image, a patch of size (1 + 2 * size) is extracted, where size is determined by setSize(int). Mean intensities of all patches are summarized in a histogram.<br /># More info: https://link.springer.com/article/10.1007/s11263-011-0430-6</p>\r\n<p>- <strong>Moments (features X3136 to X3139)</strong><br />Calculates the basic statistical moments of an image like mean, std_dev, skewness and kurtosis. The complex moments descriptors provides a complete set of invariants without any considerable calculation for rotational changes. This is achieved by using the phase of a reference component (C21, C32, ...) to cancel the effect of rotational angle variations by simple relation.<br /># More info: https://link.springer.com/chapter/10.1007/3-540-19036-8_40</p>\r\n<p>- <strong>MPEG7ColorLayout (features X3140 to X3172)</strong> <br />A color layout descriptor (CLD) is designed to capture the spatial distribution of color in an image. The feature extraction process consists of two parts; grid based representative color selection and discrete cosine transform with quantization. Color is the most basic quality of the visual contents, therefore it is possible to use colors to describe and represent an image. The MPEG-7 standard has tested the most efficient procedure to describe the color and has selected those that have provided more satisfactory results. <br /># More info: https://en.wikipedia.org/wiki/Color_layout_descriptor</p>\r\n<p>- <strong>MPEG7EdgeHistogram (features X3173 to X3252)</strong><br />This descriptor can easily generate an extra histogram bin from the 5-bin local edge histogram of each 4 × 4 sub-image. This extra histogram bin defines the ratio of the non-edge area (i.e., monotonous region) in the sub-image. <br /># More infor: https://link.springer.com/chapter/10.1007%2F978-3-540-30543-9_73</p>\r\n<p>- <strong>OpponentHistogram (features X3253 to X3316)</strong><br />The opponent histogram is a combination of three 1-D histograms based on the channels of the opponent color space. The histogram intervals for the opponent color space have ranges different from the RGB model.<br /># More info: https://www.koen.me/research/pub/vandesande-cvpr2008.pdf</p>\r\n<p>- <strong>PHOG (features X3317 to X3356)</strong><br />Generates Pyramid Histograms of Oriented Gradients (PHoG). In this descriptor, canny edge extraction is applied prior to extracting the gradients. As edge detection can be performed in several ways, it is not hard wired. <br />It can be enabled in the properties by setting features.phog.canny to true. The parameters for the canny operator are also taken from the properties file.<br /># More info: http://dl.acm.org/citation.cfm?id=1282340</p>\r\n<p>- <strong>ReferenceColorSimilarity (features X3357 to X3433)</strong><br />Color similarity features. This is not a histogram, because the reference colors are processed independently; any subset of dimensions gives the same result as computing just these colors, making this feature space very favorable for feature bagging and other projections. However, with an increasing number of colors, you can expect the dimensions to be strongly correlated.<br /># More info: http://ceur-ws.org/Vol-772/multiclust2011-paper7.pdf</p>\r\n<p>- <strong>Tamura (features X3434 to X3451)</strong><br />This descriptor is one of texture implementations and include three Tamura's features. They are coarness, contrast and directionality. <br /># More info: https://books.google.com.br/books?id=BrkMK6SA304C&amp;</p>\r\n<h2>Competition Organizers</h2>\r\n<p>Dr Ana Carolina Lorena, Computer Science Professor at Federal University of São Paulo (UNIFESP), São José dos Campos campus<br />Dr Daniel dos Santos Kaster, Computer Science Professor at Universidade Estadual de Londrina (UEL)</p>\r\n<p>Dr Elaine Ribeiro de Faria, Computer Science Professor at Universidade Federal de Uberlândia (UFU)</p>\r\n<p>Dr Francisco Carlos Rocha Fernandes, Physics Professor at UNIVAP <br />Dr Irapuan Rodrigues, Physics Professor at UNIVAP <br />Jennifer Nielsen, Aeronautics Engeenering student at UNIVAP<br />Dr Ricardo Cerri, Computer Science Professor at Federal University of São Carlos<br />Dr Vinicius Veloso de Melo, Computer Science Professor at UNIFESP – São josé dos Campos</p>\r\n<p><strong>Acknowledgements</strong></p>\r\n<p>Luis Fernando Milano Oliveira, Master Student at UEL</p>\r\n<p>Rodrigo Kuerten, student at UEL</p>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "<p>The training dataset contains 80 labeled instances (54 non-meteors and 26 <g class=\"gr_ gr_38 gr-alert gr_spell gr_inline_cards gr_run_anim ContextualSpelling ins-del\" id=\"38\" data-gr-id=\"38\">meteors</g>). The test dataset has 42 unlabeled instances (27 non-meteors and 15 <g class=\"gr_ gr_39 gr-alert gr_spell gr_inline_cards gr_run_anim ContextualSpelling ins-del\" id=\"39\" data-gr-id=\"39\">meteors</g>). Meteors are labeled as 1 and non-meteors are labeled as 0 (the class is in the last column of the dataset). The train and test datasets contain all features extracted from the images, while the zip file DatasetPerFeature.zip contains the training and testing partitions separated according to each type of characteristic extracted from the images.</p>\r\n<p>The evaluation metric for this competition is the log-loss. Therefore, the submissions must include probabilities for each class. </p>\r\n<h2>Deadline</h2>\r\n<p>The submissions will be closed on September 18th, 2017. Therefore, the final results will be based on the solutions posted until September 18th, 2017. Please note that the system will still be able to receive submissions after this deadline, but they will be not taken into account for computing the final results.</p>\r\n<h2>Award Session</h2>\r\n<p>As this competition is part of the joint activities of the 2017 editions of the Brazilian Conference on Intelligent Systems (BRACIS), the Brazilian Symposium on Databases (SBBD) and the Symposium on Knowledge Discovery, Mining and Learning (KDMiLe), the top three teams will be invited to present their solutions at a competition award session, on October 3rd, 2017 (18-19 hs). The actual final positions of the ranking will be disclosed during the conference dinner. </p>\r\n<h2>Submission Format</h2>\r\n<p>Submission files should contain three columns: Id, <g class=\"gr_ gr_48 gr-alert gr_gramm gr_inline_cards gr_run_anim Punctuation only-ins replaceWithoutSep\" id=\"48\" data-gr-id=\"48\">Prob1</g> and Prob2. Id should give the identification of the example (some Ids correspond to training examples and others are from the test examples), as indicated by the numbers in the first column of the training and testing datasets. Prob1 is the probability towards class <strong>others</strong>, while Prob2 is the probability towards class <strong>meteor</strong>. Therefore, two probabilities should be provided per example.</p>\r\n<p>For easy of reference, the Ids of the training examples are: 1, 2, 5, 8, 10, 11, 12, 15, 17, 19, 22, 24-32, 34, 36 to 45, 47, 48, 49, 52, 54, 57-62, 64, 66, 68, 70, 73, 74, 76-85, 87-90, 93, 95, 96, 98, 99, 100, 102, 104-107, 109, 110, 114, 117-120 and 122. </p>\r\n<p>The Ids of the test examples are: 3, 4, 6, 7, 9, 13, 14, 16, 18, 20, 21, 23, 33, 35, 38, 46, 50, 51, 53, 55, 56, 63, 65, 67, 69, 71, 72, 75, 86, 91, 92, 94, 97, 101, 103, 108, 111, 112, 113, 115, 116, 121.</p>\r\n<p>The file should contain a header and have the following format :</p>\r\n<pre>Id,Prob1,Prob2<br /><span style=\"font-size: 14px; line-height: 1.4em;\">3,0.9,0.1<br />4,0.8,0.3<br /></span><span style=\"font-size: 14px; line-height: 1.4em;\">etc.</span></pre>",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "<!-- Begin Base Rules -->\r\n<h3>One account per participant</h3>\r\n<p>You cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.</p>\r\n<h3>No private sharing outside teams</h3>\r\n<p>Privately sharing code or data outside of teams is not permitted. It's okay to share code if made available to all participants on the forums.</p>\r\n<h3>Team Mergers</h3>\r\n<p>Team mergers are allowed only if the teams to be merged have only one member each.</p>\r\n<h3>Team Limits</h3>\r\n<p>There is no maximum team size.</p>\r\n<h3>Submission Limits</h3>\r\n<p>You may submit a maximum of 2 entries per day.</p>\r\n<p>You may select up to 2 final submissions for judging.</p>\r\n<h3>Competition Timeline</h3>\r\n<p>Start Date: <strong>TBA</strong></p>\r\n<p>Merger Deadline: <strong>None</strong></p>\r\n<p>Entry Deadline: <strong>None</strong></p>\r\n<p>End Date: <strong>18/09/2017 00:00 AM UTC</strong></p>\r\n<p>The submissions will be closed at September 18th, 2017. Therefore, the final results will be based on the solutions posted until September 18th, 2017. As this competition is part of the joint activities of the 2017 editions of the Brazilian Conference on Intelligent Systems (BRACIS), the Brazilian Symposium on Databases (SBBD) and the Symposium on Knowledge Discovery, Mining and Learning (KDMiLe), the top three teams will be invited to present their solutions at a conference session dedicated to the competition. The actual final positions of the ranking will be disclosed during the event dinner, at October, 2017.</p>\r\n<!-- End Base Rules -->",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}