<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Dmitry Kiselev on Medium]]></title>
        <description><![CDATA[Stories by Dmitry Kiselev on Medium]]></description>
        <link>https://medium.com/@dmitry.kiselev?source=rss-5b2f6173f4aa------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*d-5AkuRIgpnSl7YkMHpXmw.png</url>
            <title>Stories by Dmitry Kiselev on Medium</title>
            <link>https://medium.com/@dmitry.kiselev?source=rss-5b2f6173f4aa------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 17:19:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dmitry.kiselev/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How does OBS virtual camera plugin work on Windows?]]></title>
            <link>https://medium.com/deelvin-machine-learning/how-does-obs-virtual-camera-plugin-work-on-windows-e92ab8986c4e?source=rss-5b2f6173f4aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/e92ab8986c4e</guid>
            <category><![CDATA[com]]></category>
            <category><![CDATA[plugins]]></category>
            <category><![CDATA[obs]]></category>
            <category><![CDATA[directshow]]></category>
            <dc:creator><![CDATA[Dmitry Kiselev]]></dc:creator>
            <pubDate>Tue, 22 Mar 2022 15:02:22 GMT</pubDate>
            <atom:updated>2022-03-22T15:02:22.692Z</atom:updated>
            <content:encoded><![CDATA[<p>The purpose of this article is to break down the work of OBS virtual camera for Windows operating system.</p><h4>Windows media back-ends</h4><p>In development of any standardized media application for Windows one faces the choice of media back-end. Most often, there are three options available: Video for Windows, DirectShow, and Media Foundation, but DirectShow normally wins over the other options. Why?</p><h4>Why DirectShow?</h4><p>Video for Windows is the oldest one. It is limited and not extensible. The real choice has always been between the old DirectShow and the new Media Foundation. Usually, the most recent technology is a go-to option, but not in the current circumstances. Media Foundation lacks a method to add virtual camera that could be discovered naturally, in the same way as other (non-virtual) devices. Windows 11 API allows this, but it is available only for the aforementioned operating system and is not as widespread as the alternative ones.</p><p>Thus, DirectShow is the only remaining option. See discussion details <a href="https://stackoverflow.com/a/35935646">here</a>.</p><h3>OBS API Overview</h3><figure><img alt="OBS logo" src="https://cdn-images-1.medium.com/max/480/1*Fc4n3TNZa1TRI9qEe1IySg.png" /></figure><p>OBS has <a href="https://obsproject.com/docs/">extensive documentation</a> including <a href="https://obsproject.com/docs/reference-modules.html">module API reference</a>.</p><p>Plugins could be made for four types of modules: sources, outputs, encoders, and services. Sources are used to render video and audio. Encoders are obviously encoders of video and audio data, implemented specifically for OBS. Outputs allow to output rendered video and audio in the forms of raw or encoded data. Services are just custom implementations for streaming services which are used in conjunction with outputs for that particular service.</p><p>For any plugin, to implement a corresponding object, you need to define a particular structure and fill it out with information and callbacks related to your plugin. You can find more information <a href="https://obsproject.com/docs/plugins.html">here</a>.</p><h3>DirectShow Overview</h3><h4>Brief summary of component object model</h4><p>COM is a binary-interface standard for software components developed by Microsoft. This means that COM could potentially be implemented on any platform.</p><p>It has a <em>very </em>extensive, almost bloated <a href="https://docs.microsoft.com/en-us/windows/win32/com/component-object-model--com--portal">documentation</a>.</p><p>What matters here is that COM objects are using interfaces to connect between each other and binary standardized interfaces only. This means that many components can interact, while being located in different processes or even on different machines.</p><p>The fact that COM has servers and clients is another notable point. Servers contain COM objects, which will be loaded and given to clients on-demand. On Windows, servers are usually executables (EXE) or dynamic-link libraries (DLL). These files can be placed anywhere on the host system, that’s why COM has concepts of a <em>registry</em> and a <em>Service Control Manager </em>(SCM). SCM locates components on local and remote hosts and connects servers to clients. And the registry exists to track where components are deployed, both on local and remote systems.</p><p>On Windows OS, the role of COM registry is usually fulfilled by the Windows registry. And before any freshly acquired component could be used, its server must be registered. On Windows, registration involves invocation of regsvr32 utility with server name, e.g.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/8e386eafb64df28657bd3204e935ab4d/href">https://medium.com/media/8e386eafb64df28657bd3204e935ab4d/href</a></iframe><h4>Brief summary of DirectShow</h4><p><a href="https://docs.microsoft.com/en-us/windows/win32/directshow/directshow">DirectShow</a> is a multimedia framework and API. It is based on COM and if one wants to use DS, he must know at least the basics of working with Component Object Model. There are several important things one needs to know about it.</p><figure><img alt="DirectShow architecture diagram" src="https://cdn-images-1.medium.com/max/589/1*nI8XggX89pSZj86VnqMS8w.png" /></figure><p>DirectShow <a href="https://docs.microsoft.com/en-us/windows/win32/directshow/introduction-to-directshow">was designed</a> for C++ only. It is based upon the concept of data processing graph creation, “filters” are its basic building blocks, implemented as COM objects. Each filter performs some kind of operation: generates, transforms or consumes video, audio and other things. Filters are chained into pipelines to perform complex data processing.</p><p>Each filter has one or more “pins” — objects that are responsible for connection between filters. When an output pin is connected to an input pin, they negotiate media types, connection parameters and if they could be connected at all.</p><p><a href="https://docs.microsoft.com/en-us/windows/win32/directshow/about-the-filter-graph-manager">Filters are always free-threaded and loaded from in-process COM servers only</a>.</p><h4>Baseclasses</h4><p><a href="https://docs.microsoft.com/en-us/windows/win32/directshow/directshow-base-classes">Base class library</a> is a sample provided with Windows SDK (or at least it <em>was</em> provided not long ago). It contains implementation of common DirectShow filter functionality. Baseclasses could be ignored entirely, but it’s not recommended. DirectShow exposes interfaces that could be implemented “from scratch”, with enough work, but not every project needs this fine control over filter behavior. Not to mention the astounding difficulty of such a task.</p><h3>Step-by-step description of how data is handled in virtual camera</h3><p>Virtual camera plugin in question is <a href="https://github.com/Fenrirthviti/obs-virtual-cam">this one</a>. To be more specific, <a href="https://github.com/Fenrirthviti/obs-virtual-cam/commit/136725a82b98d44e3d29088b90e1817625724217">this particular commit</a>.</p><h4>Preparations</h4><p>The first thing to note is that the virtual camera project has several artifacts instead of one. OBS plugin must be copied in the OBS plugin folder. DirectShow server must be copied into the appropriate location (readme recommends obs-studio install folder) and registered. Otherwise, the virtual camera plugin will not work.</p><h4>OBS Side</h4><p>The first thing that happens to every OBS plugin is <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-output/virtual_main.cpp#L14">initialization via obs_module_load function</a>. It registers two OBS plugin objects: one output <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-output/virtual_output.cpp#L245">here</a> and one source (which is named “filter”) <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-output/virtual_filter.cpp#L263">here</a>.</p><p>The main interest is the output (because virtual camera deals with the sequence “obs output → queue → directshow source”). When output starts (with virtual_output_start) a queue is created. Then, during virtual camera work, frames are pushed to the queue.</p><p>Once started, the plugin <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-output/virtual_filter.cpp#L186">will set a callback</a>, which in turn <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-output/virtual_filter.cpp#L114">will be pushing videodata</a> down the queue indefinitely, until stopped.</p><h4>Queue</h4><p>Since the OBS plugin and the DirectShow filter reside in different processes, they need some form of interprocess communication to transfer frame data from former to latter. This is where the queue comes into play.</p><p>Queue manages a file mapping, accessing it <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/queue/share_queue_write.cpp#L20">for writing</a> on the OBS side and accessing it <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/queue/share_queue_read.cpp#L10">for reading</a> on the DirectShow side. Moreover, it creates mapping with INVALID_HANDLE_VALUE, effectively mapping a file that does not exist, i.e. creating a shared memory region for two processes. <a href="https://docs.microsoft.com/en-us/windows/win32/api/winbase/nf-winbase-createfilemappinga">That memory is still backed by the system paging file though</a>.</p><p>More information about file mappings can be found <a href="https://docs.microsoft.com/en-us/windows/win32/memory/file-mapping">here</a>.</p><h4>DirectShow Side</h4><p>The DirectShow side is represented by the DirectShow filter <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-source/virtual-cam.h#L37">CVCam</a> and its pin <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-source/virtual-cam.h#L54">CVCamStream</a>. The latter is indeed a pin, because it inherits CSourceStream from baseclasses. And CSourceStream (as shown <a href="https://docs.microsoft.com/en-us/windows/win32/directshow/csourcestream">here</a>) inherits CBasePin, which in turn implements IPin interface along with many other things.</p><figure><img alt="CSourceStream class inheritance diagram" src="https://cdn-images-1.medium.com/max/347/1*clXzJ4OaxRbCwDX-u1uwaA.png" /></figure><p>This kind of complexity is precisely the reason why coding a new DirectShow filter without baseclasses is an awful pain. Without them, one would need to implement all those things by himself, relying solely on DirectShow interfaces.</p><p>The main point of interest in the DirectShow side is <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-source/virtual-cam.cpp#L147">CVCamStream::FillBuffer</a>. Because when graph setup is complete and all media types are set upon, this function does actual work pushing virtual camera frames down the graph.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/c459c1c8046fa1a917e164271e6f6575/href">https://medium.com/media/c459c1c8046fa1a917e164271e6f6575/href</a></iframe><p>FillBuffer gets any COM object implementing IMediaSample interface and fills it with frame data. First, it opens the queue, if it was not opened already. Before doing anything with the data, the function synchronizes the queue by <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-source/virtual-cam.cpp#L180">waiting</a> (if it’s too far ahead on schedule), or <a href="https://github.com/Fenrirthviti/obs-virtual-cam/blob/136725a82b98d44e3d29088b90e1817625724217/src/virtual-source/virtual-cam.cpp#L184">skipping</a> frames (if it’s too far behind). Then, if it is time to get a frame, FillBuffer pulls it from the queue. Alternatively, if there are no frames available, the function will wait for 5 milliseconds then check again. As a minor note, FillBuffer also takes care of OBS to DirectShow timestamp conversion.</p><h3>Summary</h3><p>Although the inner workings of virtual camera plugin are relatively simple, they may be difficult to grasp at first glance. Many things here are done behind the scenes, just because base class library is thought out this way. We hope that this short description will be of help to anyone who wants to understand the inner workings of OBS virtual camera plugin on Windows or to implement an analogue.</p><p>Read more articles on Machine Learning from our <a href="https://medium.com/deelvin-machine-learning">Deelvin Machine Learning Blog</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e92ab8986c4e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/deelvin-machine-learning/how-does-obs-virtual-camera-plugin-work-on-windows-e92ab8986c4e">How does OBS virtual camera plugin work on Windows?</a> was originally published in <a href="https://medium.com/deelvin-machine-learning">Deelvin Machine Learning</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning  SDK]]></title>
            <link>https://medium.com/deelvin-machine-learning/machine-learning-sdk-edc4c23824f0?source=rss-5b2f6173f4aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/edc4c23824f0</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[software-development-kit]]></category>
            <category><![CDATA[image-segmentation]]></category>
            <dc:creator><![CDATA[Dmitry Kiselev]]></dc:creator>
            <pubDate>Tue, 12 Jan 2021 15:01:07 GMT</pubDate>
            <atom:updated>2021-01-12T15:01:07.449Z</atom:updated>
            <content:encoded><![CDATA[<h3>Machine Learning Software Development Kit: an introduction and main functions</h3><p>Today our <a href="https://deelvin.com/">Deelvin</a> team would like to present Machine Learning Software Development Kit (ML SDK). It is a library which takes care of any ML related problems that one would need to solve. The first version of C API is somewhat simple, if not primitive, but it will undergo further development in the future.</p><p>ML SDK is a product to solve tasks in the field of computer vision.</p><p>Currently, ML SDK allows one to:</p><ul><li>Detect ads in video streams or any frame sequences</li><li>Detect or perform segmentation of video compression algorithm artifacts</li><li>Perform human image semantic segmentation (and you can watch our demo <a href="https://medium.com/deelvin-machine-learning/background-remover-is-now-updated-to-version-1-1-c1de8a97df9b">here</a>)</li><li>Upscale low resolution images using machine learning</li><li>Detect various objects and their coordinates.</li></ul><p>This article serves two purposes: it is a press release and a short manual of the product. The article will demonstrate how ML SDK works and describe all its functions.</p><p>The following notes describe a general API structure:</p><ol><li>Each of the above problems has a “module” related to it and each such module has a short name, which serves as a prefix. For example, perseg is the prefix for a person segmentation module.</li><li>Usually, for any kind of user-controlled object there exist two functions: &lt;module-name&gt;_*_create_&lt;object&gt; and &lt;module-name&gt;_*_delete_&lt;object&gt; , with obvious intended purpose.</li><li>For “detectors”, “segmentators” and alike there are &lt;module-name&gt;_*_predict functions. They provide an interface to utilize model inference with some convenient pre- and post-processing.</li><li>Every function works synchronously. Optionally, there may be several functions named &lt;module-name&gt;_*_async which provide asynchronous interface to their counterparts (with the use of user callbacks).</li><li>ML SDK uses its own structures as input and output to functions, but they try to mimic a familiar layout. For example, an image is made like frame planes from AVFrame in ffmpeg, plus two user fields.</li></ol><h3>Showcase of work with ML SDK</h3><p>As an example, work with person segmentation module will be demonstrated. More detailed info about human image semantic segmentation can be found <a href="https://medium.com/deelvin-machine-learning/human-image-segmentation-experience-from-deelvin-5148a6cc71da">here</a>.</p><pre>// Create segmentator and check if creation has failed<br>CVQPersonSegmentator segmentator = perseg_create_segmentator();<br>if (!segmentator) {<br>    fprintf(stderr, &quot;%s\n&quot;, perseg_errmsg());<br>    exit(-1);<br>}</pre><pre>// Create Image struct and fill it<br>struct Image image;<br><br>image.image_data.width = 1920;<br>image.image_data.height = 1080;<br>int32_t stride[1] = {1920 * 3};<br>image.image_data.stride = stride;<br>image.image_data.colorspace = PIXEL_FORMAT_RGB24;<br><br>// Raw unformatted RGB24 bytes<br>// This buffer is defined and loaded somewhere else<br>//uint8_t* buffer[1] = {(uint8_t*)malloc(1920 * 1080 * 3)};<br>image.image_data.buffer = buffer;</pre><pre>// Just create mask, no need to think about its structure or what exact size is needed<br>struct Mask* mask = perseg_create_mask(image.image_data.width, image.image_data.height);<br><br>// Prediction<br>// Notice that user owns all arguments<br>// Nothing is allocated inside and is needed to be cleaned up later<br>perseg_predict(segmentator, &amp;image, mask);</pre><pre>// But since user owns allocated storage, user must take care of it<br>//free(buffer[0]);<br>perseg_delete_mask(mask);<br>perseg_delete_segmentator(segmentator);</pre><p>It’s a snippet from a person segmentation code sample with all unnecessary stuff omitted. A bit of a warning is in place: ML SDK <em>does not</em> support any kind of codec or picture format out-of-the-box. It could be done purposefully, though.</p><p>The images below demonstrate how ML SDK functions. ML SDK takes the left image as an input and produces person segmentation output visualized with the right image. In the second image the person’s contours are clearly outlined in white against the black background. ML SDK did a good job separating the hair as well as all other details of the image.</p><figure><img alt="Person segmentation output visualization" src="https://cdn-images-1.medium.com/max/1024/1*jm2uOa5OK73xuuHpj2jlsQ.jpeg" /><figcaption>Person segmentation output visualization</figcaption></figure><h3>A short summary and the list of all functions</h3><h4>Content detection (vqcd)</h4><p>Content detection takes one image from a sequence as its input and it can return structure with a predicted type of content (payload/ad) and a number of frames that have been fallen into that type. It also can silently eat frames until it will be sure about the content type and the number of frames in the current batch.</p><p>The functions vqcd_create_detector and vqcd_delete_detector obviously create and delete this detector. vqcd_predict is there to synchronously process next frame in sequence, but it may do nothing. In addition to output with info, there is a boolean value that says if current image has triggered inference. If vqcd_predict returned false , then output is meaningless (and was not touched at all).</p><p>This element has asynchronous mode of work. vqcd_init_async and vqcd_abort_async can be used to start and stop separate thread (it also stops automatically when detector is destroyed). vqcd_predict_async is an analogue of vqcd_predict , but it will use user callback provided to vqcd_init_async to notify the user about its prediction. vqcd_lock_context_async and vqcd_unlock_context_async , as one can infer from their names, can be used to lock and unlock user context mutex.</p><h4>Artifact detection/segmentation (artdet)</h4><p>Originally, this module was intended to handle only artifact detection, but later it deemed reasonable to have segmentation as well. Now it does only the latter, which can also be used to emulate the former, albeit slower.</p><p>There are several segmentators in this module. The first is a simple one. It takes an image as its input and returns a mask (one channel / grayscale image) to highlight every artifact it has found in the image.</p><p>The functions artdet_create_segmentator and artdet_delete_segmentator create and delete it, respectively. artdet_predict is there to run inference. artdet_create_mask should be used to create an appropriate mask, with artdet_delete_mask to clean it up later. artdet_errno and artdet_errmsg could be used to get an idea of which kind of error has occurred.</p><p>The second kind of segmentation is able to pinpoint different types of visual artifacts. It returns separate masks for noise, block and blur artifacts.</p><p>Again, artdet_multiple_create_segmentator and artdet_multiple_delete_segmentator manage the element. artdet_multiple_predict uses inference. artdet_multiple_create_mask_array and artdet_delete_mask_array manage the mask array, which will be the result of computation. artdet_multiple_errno and artdet_multiple_errmsg get a hang of a possible error.</p><p>Last, but not least, there are three segmentators that produce mask arrays for artifacts specific to a particular algorithm type: AVC, MJPEG, Wavelett or HEVC. Their main differences are in the accuracy aspects and existence or lack of probability labels.</p><p>Several other important functions are the following.</p><p>To create and delete:</p><ul><li>artdet_multiple_2_create_segmentator and artdet_multiple_2_delete_segmentator</li><li>artdet_multiple_w_labels_create_segmentator and artdet_multiple_w_labels_delete_segmentator</li><li>artdet_multiple_w_labels_2_create_segmentator and artdet_multiple_w_labels_2_delete_segmentator</li></ul><p>To make a prediction:</p><ul><li>artdet_multiple_2_predict</li><li>artdet_multiple_w_labels_predict</li><li>artdet_multiple_w_labels_2_predict</li></ul><p>To manage mask arrays:</p><ul><li>artdet_multiple_2_create_mask_array</li><li>artdet_multiple_w_labels_create_mask_array</li><li>artdet_multiple_w_labels_2_create_mask_array</li><li>artdet_delete_mask_array</li></ul><p>To understand errors:</p><ul><li>artdet_multiple_2_errno and artdet_multiple_2_errmsg</li><li>artdet_multiple_w_labels_errno and artdet_multiple_w_labels_errmsg</li><li>artdet_multiple_w_labels_2_errno and artdet_multiple_w_labels_2_errmsg</li></ul><h4>Person segmentation (perseg)</h4><p>It takes an image as its input, and gives a mask, which highlights all humans on that input picture. The mask format may seem to be an unusual one: it’s 0x00&#39;s and 0x01’s only.</p><p>The interface of this module is pretty obvious and straightforward. perseg_create_segmentator and perseg_delete_segmentator to manage segmentator. perseg_create_mask and perseg_delete_mask to manage masks. perseg_predict for inference. perseg_errno and perseg_errmsg to inspect errors that have occurred.</p><h4>Video super resolution (VSR)</h4><p>VSR takes an image as input, and gives an image as output. Its purpose is to obtain a picture with higher resolution using machine learning. The current ratio is 4:1 vertically and horizontally. It means that the output image will be 16 times bigger.</p><p>The interface of VSR is simple too. vsr_create_upscaler and vsr_delete_upscaler to create or delete upscaler. vsr_alloc_upscaled_image and vsr_delete_upscaled_image to manage upscaled image. Upscaled image is based on input one, thus user should not be bothered with calculations of size and memory allocation. vsr_errno and vsr_errmsg is there to identify an error cause.</p><h4>Object detection (objdetect)</h4><p>This one detects objects. It takes an image as its input and returns a variable amount of boxes to the user. Each box being a four number designation of a rectangle (x, y, width, height) with a class label on it.</p><p>Also, because of unknown number of boxes, this module is the only one that allocates memory on heap and dumps control of it onto the user.</p><p>objdetect_create_detector and objdetect_delete_detector are managing the detector. With objdetect_create_box_array and objdetect_delete_box_array to manage array of boxes and the user will not notice that something is off, but it is. One of the box array structure fields is assigned to a memory address allocated by the library. objdetect_predict is there to use inference. objdetect_errno and objdetect_errmsg to look up info about the error, should one occur.</p><h3>Conclusion</h3><p>Right now ML SDK is available as a trial version upon request. It is undergoing further development and it is expected that its API may change. Follow its development on <a href="https://deelvin.com/">Deelvin website</a>. You can see ML SDK live using <a href="https://deelvin.com/ml-player">ML Player</a>. For inquiries please contact our <a href="https://deelvin.com/contact-deelvin">Deelvin team</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=edc4c23824f0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/deelvin-machine-learning/machine-learning-sdk-edc4c23824f0">Machine Learning  SDK</a> was originally published in <a href="https://medium.com/deelvin-machine-learning">Deelvin Machine Learning</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>