[{"sys":"1","total":2,"skip":0,"limit":12,"items":"2","includes":"3"},{"type":"4"},["5","6"],{"Entry":"7","Asset":"8"},"Array",{"sys":"9","fields":"10"},{"sys":"11","fields":"12"},["13","14","15","16","17","18","19","20","21","22","23","24","25"],["26","27","28","29"],{"space":"30","id":"31","type":"32","createdAt":"33","updatedAt":"33","environment":"34","revision":1,"contentType":"35","locale":"36"},{"title":"37","teaserImage":"28","teaserSummary":"38","body":"39","author":"22","date":"40","relatedArticles":"41","tagsCloud":"42","category":"13"},{"space":"43","id":"44","type":"32","createdAt":"45","updatedAt":"46","environment":"47","revision":5,"contentType":"48","locale":"36"},{"title":"49","teaserImage":"29","teaserSummary":"50","body":"51","author":"17","date":"52","tagsCloud":"53","category":"13"},{"sys":"54","fields":"55"},{"sys":"56","fields":"57"},{"sys":"58","fields":"59"},{"sys":"60","fields":"61"},{"sys":"62","fields":"63"},{"sys":"64","fields":"65"},{"sys":"66","fields":"67"},{"sys":"68","fields":"69"},{"sys":"70","fields":"71"},{"sys":"72","fields":"73"},{"sys":"74","fields":"75"},{"sys":"76","fields":"77"},{"sys":"78","fields":"79"},{"sys":"80","fields":"81"},{"sys":"82","fields":"83"},{"sys":"84","fields":"85"},{"sys":"86","fields":"87"},{"sys":"88"},"4W0sXygh1f6A3v5MstNhND","Entry","2020-06-23T09:51:27.783Z",{"sys":"89"},{"sys":"90"},"nl","Action Recognition, the science of detecting human interactions","A blog about human action recognition. How does this work and what models can you use to detect different movements or gestures?","__With the increasing popularity and demand of deep learning applications many new interesting use cases have been developed. Think of applications that offer emotion detection, age prediction, object recognition and many more possibilities. A field you might not have heard about yet is called (human) action recognition, the art of detecting specific actions or gestures. In this blog I take a closer look at the different detection models available for action recognition and talk about my smart shelf experiment.__\n\n![LivelyIdleArgali-mobile](//images.ctfassets.net/w4dg3cjf42ew/63zvBPvBL0DD9KoNqJ6taF/f1dae1dab304e8ddd4ecda53558ceab3/LivelyIdleArgali-mobile.gif)\n\n### Action recognition\nThe human body is capable of making many different movements and gestures. It is still a challenge to detect them all using computers. The field of study that is concerned with classifying this, is called human action recognition. These classifications are based on data which can be obtained from sensors or cameras. The difficulty in classifying these movements lies in the variety of movements made by a person and the environment in which these movements are recorded. As an example, an environment with too much light or background noise can lead to occlusion in the image, which leads to faulty classifications. \n\nThese varieties make the task of classification more difficult, sometimes very complex! Not only is the variety or occlusion a challenge, but the lack of computational power can also be problematic. Especially if you want to process captured information during every frame of a video.\n\n### Deep learning networks\nLuckily there are solutions for that, they come in the form of pre-trained deep learning networks and object detection models. These ready-to-use networks are primarily optimized for computers with less compute power and are easily accessible for everyone. Thanks to these models we don’t have to go to the complex challenge of training our own deep learning network. \n\nBelow an overview of some of the available deep learning networks. The models are placed inside the table based on their accuracy (visible on the x-axis) and number of operations. For now just focus on the accuracy. \n\n![Picture 1](//images.ctfassets.net/w4dg3cjf42ew/2PIXlAbQDVDv8U2yrf2eOQ/fddb9b4571c3d6d7ab2a2c4255febb6e/Picture_1.png)\n*Available deep learning models and their accuracy scores ([source](https://medium.com/@culurciello/analysis-of-deep-neural-networks-dcf398e71aae))*\n\n### Detection models\nThe table below contains an overview of some of the most commonly used object detection models. The first column displays the model name and the top row indicates the dataset the model was trained and tested on. COCO and PASCAL VOC are the two most common used datasets for training. \n\n![Picture 2](//images.ctfassets.net/w4dg3cjf42ew/6hBLu1u7LjyVmHo28AGciU/cc914441d425bb032b3dfd18c3d3c758/Picture_2.png)\n*Overview of the mAP scores on the 2007, 2010, 2012 PASCAL VOC dataset and 2015, 2016 COCO datasets ([source](https://medium.com/zylapp/review-of-deep-learning-algorithms-for-object-detection-c1f3d437b852)).*\n\nFor my experiment I chose to use the MobileNet model in combination with an SSD detection model (highlighted blue in the table above). Both models have a trade-off in accuracy and speed I found out. Besides, MobileNet is designed for edge devices, such devices can be Raspberry Pi’s, smartphones, or any device with limited power. A great advantage for low budget experiments.\n\n### Experiment – tracking consumer behavior\nI originally tried to create a smart shelf which can be used to monitor consumer actions inside a store, providing retail store managers new actionable insights. \nTo achieve this, I try to gain insights of customer demographics, better known as face expressions. Think of face expressions you show when looking at a product or commercial.\n\nBelow a small overview of the data that’s being captured:\n\n-\tFace tracking (follow the face through the video stream)\n-\tHand movement\n-\tFace metrics (emotions, such as joy, surprise, anger, sadness…)\n-\tFace landmarks (important facial structures, jaw, eyes, nose…)\n-\tNumber of people in front of the shelf\n-\tDwell time (How long spends a person in front of the shelf)\n\nTake a look at the video below which shows what the action recogniton model is capable of to do in this stage of the project. Thoroughly testing with different persons however is a challenge due to the currrent #STAYATHOME situation. This lets the opportunity rise for our creative minds to find alternative ways of testing. I simply use my phone and play a video containing multiple persons and hold the device in front of the camera. It is simple, but it works ;)\n\nAction Recognition\n\n\n### New opportunities\nWith an indefinitely stay at home society, the original purpose of the smart shelf experiment slowly fades away. This absolutely does not mean that the experiment is a waste of has failed. This “problem” brings the opportunity to find new means and implementations for the smart shelf. The different components used for my smart shelf experiment are designed in such a generic way that they can easily be implemented for other use cases. Because after all, it just captures data and depending on the data you can do with it whatever you want. Here are some idea’s I have about implementation outside retail:\n\n-\t__Mood detector__. Detecting your mood at home and change the lights or music accordingly\n-\t__Self-serving restaurants__. The shelf would keep track what a customer puts on his/her tray\n-\t__Smart storage__. Keep track of products in storage shelves (at home or restaurants) and automatically add a product to the shopping list when its minimum threshold has been exceeded. \n-\t__Social distance checker__. Detect if people follow the 1.5m distance guideline.\n\n### Next steps\nThe next step would be to incorporate action recognition. This could lead to a function in which the smart shelf can detect if a customer intends to buy a product or is trying to steal something. Additional features such as face mask detection could also be implemented. This could assist in checking for mandatory wear of face masks. \n\n\n","2020-06-23T00:00+02:00",["16","23"],["15","24","20","25","21"],{"sys":"91"},"29A8JeV5ahJIDXZsUG0FpN","2020-05-08T07:38:15.858Z","2020-05-08T09:24:00.055Z",{"sys":"92"},{"sys":"93"},"Train your presenting skills in Virtual Reality","Explore & learn in a different way using Virtual Reality. Read more about this modern learning approach in this blog story.","__The current COVID-19 pandemic has a huge social impact, we work from home and communicate with each other in a more digital way then before. This rare and difficult situation also brings a lot of creativity, e.g. new online collaborative tools are launched, and we experience new ways to connect since we are not allowed to meet face to face anymore. Distance learning for example really fits in, as we are already familiar with online courses nowadays. With virtual reality we can improve the experience even further, read more about our VR training solution in this blog post.__\n\n![BGN0872 2[1]](//images.ctfassets.net/w4dg3cjf42ew/obx1IcApt4s0HxrSr8sSP/ac82bb48e64bb21d756fa8809dec1a56/_BGN0872_2_1_.jpg)\n\nTalking about modern learning, VR really fits in. It allows us to explore & learn in a different way. In his [blogpost](https://julianstodd.wordpress.com/2020/01/29/virtual-reality-a-lens-for-learning/), Julian Todd discusses a framework for learning using VR. As he states *“VR scenarios enable us to manipulate the world around us (or even other worlds), in ways that transcend reality: consider ‘Physics and Space’, ‘Problem Solving and Creativity’, and ‘Rehearsal and Prototype’. In VR worlds things like friction, weight, and mass, become a matter of choice, not of fact. We can creatively interact without risk or impact.”*\n\nBesides, for us the starting point to create a VR learning studio was the practical challenge to execute class-room training for a lot of our colleagues. Also consider, when training your presentation skills, analyzing recorded videos is although an effective methodology, a lot of companies have sufficient facilitators across the globe to deliver a two-day classroom session. So we designed a solution that builds the capability (using reflective learning methodology) and at the same time doesn’t take much time to use.\n\n### Virtual Reality & AI\nWe decided to create a cost-effective solution using a VR based simulation environment, where people gain the knowledge of how to present and also enjoy the opportunity to present in a non-threatening environment. Using an HTC Vive, Oculus Quest or Android Mobile device you can learn presenting in a safe environment like your own home for example.\n\nAnother great advantage of using these devices is that we are able to provide instant feedback on about 20 parameters around presentation skills and even storytelling skill using natural language processing (NLP).\n\n![Screenshot 2020-05-08 at 10.09.04](//images.ctfassets.net/w4dg3cjf42ew/2nWtB3kIf21qts6i13FvMC/d09e6b840baf9ee7ce771a695435aa61/Screenshot_2020-05-08_at_10.09.04.png)\n\nSo how does this work in practice? After putting on the VR headset and (optional) upload your own PowerPoint presentation the application will take a sample of your voice as a baseline for further analysis. As a next step the student will enter a virtual meeting room where he starts presenting as you would expect in a real-life situation. During the presentation the application captures all kind of data like audio, speech and physical information like eye contact and the movement of your head. With this it can analyse for example if you’re looking at the audience and the different attendees. \n\n![Screenshot 2020-05-08 at 10.09.21](//images.ctfassets.net/w4dg3cjf42ew/7ciJpslqjGKd4m0afriYLm/d0ab62120d9bd39ece79839bc5911201/Screenshot_2020-05-08_at_10.09.21.png)\n\nOther examples of presentation skill measurement are audio analysis – like volume and pitch and in depth speech analysis. This includes the speed of your voice, the amount of filler words (ehh ah so), your tone of voice and emotion. Voice emotion analysis – using cognitive cloud services – gives insights in your emotion flow during the presentation, e.g. neutral, happy, fear, stress, etc.\n\n> Train your presenting skills using VR in a safe environment like your own home\n\nvr academy room\n\n\n\n### Storytelling analysis\nAnother interesting development is the ability to analyse your pronunciation. Determining words, phrases and sentences to conclude if you are telling a great story, convince the audience or formulate the right way. This sound like magic and – although in continuous development –does a good job using latest AI NLP services from Google. However, at the moment only available using a pre-defined presentation and in English.\n\n### The final verdict\nWhen finishing your session results are displayed on a dashboard in the virtual world. The dashboard contains an overview of different parameters like speed of speech, use of meaningless words (or fillers), statistics like totals used words, time, volume, etcetera.\n\nvr academy results\n\n\n### To conclude\nCombining virtual reality with data analysis and AI services like NLP and voice emotion detection gives some great opportunities in self-study. Analysing soft skills without the hassle of classroom training or creating videos of your presentation skills which need to be analysed afterwards by a mentor. Our VR Academy gives real-time feedback in a safe environment, using advanced analytics and AI to give exact results, easy to retrain and keep improving on.\n\nIt is a great and fun VR experience, showing the power of todays AI capabilities and future way of modern learning possibilities.\n\n\n---\n\n\nCurious about our VR training solutions? Please contact Ankur Bhatt.\n\n","2020-05-08T00:00+02:00",["15","25","19","18","14","21"],{"space":"94","id":"95","type":"32","createdAt":"96","updatedAt":"96","environment":"97","revision":1,"contentType":"98","locale":"36"},{"title":"99"},{"space":"100","id":"101","type":"32","createdAt":"102","updatedAt":"102","environment":"103","revision":1,"contentType":"104","locale":"36"},{"title":"105"},{"space":"106","id":"107","type":"32","createdAt":"108","updatedAt":"108","environment":"109","revision":1,"contentType":"110","locale":"36"},{"title":"99"},{"space":"111","id":"112","type":"32","createdAt":"113","updatedAt":"114","environment":"115","revision":15,"contentType":"116","locale":"36"},{"title":"117","body":"118","author":"119","date":"120","tagsCloud":"121","category":"13"},{"space":"122","id":"123","type":"32","createdAt":"124","updatedAt":"125","environment":"126","revision":2,"contentType":"127","locale":"36"},{"name":"128","eMail":"129","avatar":"27"},{"space":"130","id":"131","type":"32","createdAt":"132","updatedAt":"132","environment":"133","revision":1,"contentType":"134","locale":"36"},{"title":"135"},{"space":"136","id":"137","type":"32","createdAt":"138","updatedAt":"138","environment":"139","revision":1,"contentType":"140","locale":"36"},{"title":"141"},{"space":"142","id":"143","type":"32","createdAt":"144","updatedAt":"144","environment":"145","revision":1,"contentType":"146","locale":"36"},{"title":"147"},{"space":"148","id":"149","type":"32","createdAt":"150","updatedAt":"150","environment":"151","revision":1,"contentType":"152","locale":"36"},{"title":"153"},{"space":"154","id":"155","type":"32","createdAt":"156","updatedAt":"156","environment":"157","revision":1,"contentType":"158","locale":"36"},{"name":"159","avatar":"26"},{"space":"160","id":"161","type":"32","createdAt":"162","updatedAt":"163","environment":"164","revision":6,"contentType":"165","locale":"36"},{"title":"166","body":"167","author":"17","date":"168","tagsCloud":"169","category":"13"},{"space":"170","id":"171","type":"32","createdAt":"172","updatedAt":"172","environment":"173","revision":1,"contentType":"174","locale":"36"},{"title":"175"},{"space":"176","id":"177","type":"32","createdAt":"178","updatedAt":"178","environment":"179","revision":1,"contentType":"180","locale":"36"},{"title":"181"},{"space":"182","id":"183","type":"184","createdAt":"185","updatedAt":"185","environment":"186","revision":1,"locale":"36"},{"title":"159","file":"187"},{"space":"188","id":"189","type":"184","createdAt":"190","updatedAt":"190","environment":"191","revision":1,"locale":"36"},{"title":"128","file":"192"},{"space":"193","id":"194","type":"184","createdAt":"195","updatedAt":"195","environment":"196","revision":1,"locale":"36"},{"title":"197","file":"198"},{"space":"199","id":"200","type":"184","createdAt":"201","updatedAt":"201","environment":"202","revision":1,"locale":"36"},{"title":"203","file":"204"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"211"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"211"},{"sys":"212"},"1eeYevskvmKWuggqW0Osum","2017-02-02T13:53:30.472Z",{"sys":"213"},{"sys":"214"},"Innovation",{"sys":"215"},"2L7aRIXN7PtzjEYJAnsQJl","2020-05-07T13:27:32.110Z",{"sys":"216"},{"sys":"217"},"Virtual Reality",{"sys":"218"},"2SjbJhA9va4ceCisiiQi8G","2016-06-22T09:01:54.823Z",{"sys":"219"},{"sys":"220"},{"sys":"221"},"3J1Z5sBaE4WmWoQJ2YTcCa","2019-02-20T10:45:42.222Z","2019-06-11T11:01:58.335Z",{"sys":"222"},{"sys":"223"},"Experimenting with gesture detection","__Because we believe that people will interact differently with digital touchpoints in the future, it is important to experiment, to find out what works and what doesn't.\nFor example: what if you could have an interaction with a mirror without touching it or pressing specific buttons. Within our innovation lab we are exploring several ways to detect gestures by using a simple webcam.__ \n\nMagic Mirror\n\n\nFrom a technical point of view the challenge is clear: what kind of technology or AI model do we have to use for proper gesture detection? where from a more human point of view the challenge is that people move very differently. So, we defined three approaches to find out what works best:\n\n1. Pose estimation data and using some math to calculate the performed gesture\n2. Pose estimation data using deep learning to create a model that detects specific gestures\n3. Action Recognition approach by using machine learning to detect specific actions by analyzing video frames\n\n### Approach 1: Posture detection with some math\n\nUsing Pose detection with a [Tensorflow](https://www.tensorflow.org/) based version called [Tf-pose-estimation](https://github.com/ildoonet/tf-pose-estimation) we are able to detect human bodies, hands, faces and foot keypoints in a single image or frame. When processing a video stream Tf-pose-estimation constantly provides you with coordinates of the different body parts resulting in a stream of movement data.\n\nTo detect gestures – like a swipe movement – the idea is to process the Open Pose data and detect specific gestures using a few algorithms. \n\nTo process we simply compare all the x- and y-coordinates of the ‘swipe-hand’ with pre-stored swipe gesture data. Of course, it is a bit more complicated because of the timestamp data to detect speed of movements for example.\n\nHowever, the disadvantage of this approach is the difficulty to detect deviations. A specific gesture should generate more or less the same data points as our math model expects, otherwise the gesture won’t be detected at all. This method won’t support very natural movements but more fixed movements that follow an exact path.\n\n### Approach 2: Deep learning with pose estimation data\n\nSo we decided to explore deep learning approaches. The idea behind deep learning is that a model can learn by itself by looking at large amounts of data. In the best case, the gesture model is both accurate and fast, running real-time. On practice, however, deeper models that use richer data perform better, which comes at the cost of speed. That is why we started with a simple version of the model that just uses pose estimation data. \n\nWe went for a well-known recurrent [LSTM model](https://en.wikipedia.org/wiki/Long_short-term_memory) and trained it on pre-recorded swipe movement data. We asked a few volunteers in the office to do the swipes which were stored as skeletons (much lighter compared to RGB input). To improve the model, we also recorded movements that could be confused for a swipe (e.g. taking a phone, raising hand or waving to someone). Altogether our training set consisted of labeled data with 3 gesture classes (left swipe, right swipe and no swipe).\n\n![Results Open Pose data Tensorflow](//images.ctfassets.net/w4dg3cjf42ew/4c7MfxEsoLKyAXcOWLkbK0/0589958756a146bc3ae3c48f220ab371/Results_tensorboard.png)\n\nThe results of the deep learning approach are promising as the model achieves up to 90% validation accuracy on the pre-recorded clips of gestures. To launch the model on real-life stream, however, we want to experiment with more data input types and architectures. In that field it is promising to combine recurrent models with convolutional models running on video frame data, which may still achieve real-time performance. \n\n### Approach 3: Deep learning with video data\n\nAnother promissing approach is action recognition which is about the recognition of different actions from a video file or stream. By analyzing frames as a specific action – like a swipe gesture – may or may not be performed. Read more about it in this great [blog article](http://blog.qure.ai/notes/deep-learning-for-videos-action-recognition-review).\n\nAlthough this approach costs a lot more in terms of computing power the results might be more accurate. So we decided to continue our gesture detection journey using this approach. Please stay tuned for new updates!\n\n",{"sys":"224"},"2019-03-01T00:00+01:00",["15","20","24"],{"sys":"225"},"3WyvGIHzrOyMK0QQQ2egwm","2017-02-10T09:38:57.749Z","2017-02-13T14:21:41.160Z",{"sys":"226"},{"sys":"227"},"Hayo Rubingh","hrubingh@mirabeau.nl",{"sys":"228"},"4C3pGiHyjNMfBpY8wtVP3t","2019-05-01T11:27:18.289Z",{"sys":"229"},{"sys":"230"},"Voice",{"sys":"231"},"4UvDwQdC1i0k0qUU6QWMY","2017-04-14T08:18:03.582Z",{"sys":"232"},{"sys":"233"},"Unity3D",{"sys":"234"},"4c7IhAUIxqa2OYaGGUQuwQ","2018-04-12T09:02:59.857Z",{"sys":"235"},{"sys":"236"},"AI",{"sys":"237"},"5GXp2x6769IsUg8xf65O8m","2020-05-07T13:27:50.256Z",{"sys":"238"},{"sys":"239"},"Emotion detection",{"sys":"240"},"6KPZiYMj62T4Cjsrx70iC9","2020-06-15T08:28:48.483Z",{"sys":"241"},{"sys":"242"},"Daniël Jrifat",{"sys":"243"},"7HEKUeUs00gSWYqe6iKwsO","2017-08-01T12:36:07.320Z","2019-12-17T13:46:30.751Z",{"sys":"244"},{"sys":"245"},"Our face recognition demo keeps evolving: a Summer update","Earlier this year [we wrote about](http://blog.mirabeau.nl/nl/articles/real-time_face_recognition_with_cloud_ai_services/6hhU0O1852uOAO22eKuI4g) our experiments with Cloud Cognitive Services like text-, speech-, and image recognition. This research has led to a lot of new insights, which will be published in the very near future.\n\n### A warm welcome to our offices\nOverwhelmed by the possibilities of modern cloud cognitive services, we developed a face recognition appliance. When entering our head office in Amsterdam, visitors will notice a big screen with a camera. Once a visitor’s face has been detected, further analysis takes place using cloud face recognition services based on the power of machine learning. The service provides insights like age, gender, emotion and other facial features like reading glasses, beard and moustache. A very fun and appealing demo.\n\n![Mirabeau-Face-Recognition 1080-570-pixels-256-kleuren](//downloads.contentful.com/w4dg3cjf42ew/34HOgDoovuqgeOcwkWUQQA/d63b7f58a62befa7f340818c597a0eab/Mirabeau-MarCom-Face-Recognition_1080-570-pixels-256-kleuren.gif)\n\n### Taking face recognition to the next level\nIn the last two months, new features and a fresh design were implemented to make the setup more mature and appealing. One of the improvements we made is the animated welcome message that provides a warm welcome to people entering our Mirabeau offices in Amsterdam and Eindhoven. Not in the Netherlands? The demo is also installed in the Cognizant Collaboratories in London, New-York and even Melbourne. \n\n![Mirabeau-Face-Recognition WELCOME SCREEN 1080x570-pixels-256-kleuren](//downloads.contentful.com/w4dg3cjf42ew/2Z9K4DdDDOck6I4KkG0Aiq/168c311ae8472a591d674b15303a86e8/Mirabeau-MarCom-Face-Recognition_EMPTY_SCREEN_1080x570-pixels-256-kleuren.gif)\n\nOther improvements include the ability to detect new facial features like lip & eye makeup and a look-a-like detection feature. The appliance shows a colleague you look most similar to if there is more than a 50% match.\n\n### Bringing your data to life\nThe demo shows the true power of today’s Cognitive Cloud Services powered by AI and can easily be enriched with your own data. Several use-cases can be thought of such as supporting your employees with relevant data, like their next appointment. Attracting visitors to your booth at an event is another great example. And because the system now is capable of logging statistical data, you could provide insights like the happiness level per day, top visitors, average age, etcetera.\n\n### Face recognition demo for your industry?\nAt the moment we are working on industry specific variants for our clients. Interested in bringing innovation to life in your own office? Please get in touch with Hayo Rubingh, hrubingh@mirabeau.nl.\n\n","2017-08-01T00:00+02:00",["25","15","246"],{"sys":"247"},"NNMAa681KSc8QK06AAc2I","2015-08-20T11:36:14.096Z",{"sys":"248"},{"sys":"249"},"Machine Learning",{"sys":"250"},"Zm0shwkpAAEoaeSoSoYio","2017-02-02T13:52:49.823Z",{"sys":"251"},{"sys":"252"},"Cognitive Services",{"sys":"253"},"15AxS1rNKE7opB64pWRsk1","Asset","2020-06-15T08:28:35.442Z",{"sys":"254"},{"url":"255","details":"256","fileName":"257","contentType":"258"},{"sys":"259"},"4vhOrQoaykE6GIii4m8u4K","2017-02-10T09:38:46.481Z",{"sys":"260"},{"url":"261","details":"262","fileName":"263","contentType":"258"},{"sys":"264"},"63zvBPvBL0DD9KoNqJ6taF","2020-06-15T06:44:12.963Z",{"sys":"265"},"LivelyIdleArgali-mobile",{"url":"266","details":"267","fileName":"268","contentType":"269"},{"sys":"270"},"obx1IcApt4s0HxrSr8sSP","2020-05-08T08:07:32.756Z",{"sys":"271"},"BGN0872 2[1]",{"url":"272","details":"273","fileName":"274","contentType":"258"},"Link","Space","w4dg3cjf42ew","master","Environment","ContentType","6pHF2p1ah2Aag4KAkCK6iG",{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"275"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"211"},{"type":"205","linkType":"32","id":"277"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"278"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"278"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"211"},{"sys":"279"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},{"type":"205","linkType":"210","id":"276"},{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},"//images.ctfassets.net/w4dg3cjf42ew/15AxS1rNKE7opB64pWRsk1/a5b706ac7a29c33d9fa973c3fb722973/WhatsApp_Image_2020-06-15_at_10.16.36.jpeg",{"size":55795,"image":"280"},"WhatsApp_Image_2020-06-15_at_10.16.36.jpeg","image/jpeg",{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},"//images.ctfassets.net/w4dg3cjf42ew/4vhOrQoaykE6GIii4m8u4K/718bede18df921f3ad589dd7ee7b2f33/P1070053_FINAL__3_.jpg",{"size":289686,"image":"281"},"P1070053_FINAL (3).jpg",{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},"//images.ctfassets.net/w4dg3cjf42ew/63zvBPvBL0DD9KoNqJ6taF/f1dae1dab304e8ddd4ecda53558ceab3/LivelyIdleArgali-mobile.gif",{"size":2653975,"image":"282"},"LivelyIdleArgali-mobile.gif","image/gif",{"type":"205","linkType":"206","id":"207"},{"id":"208","type":"205","linkType":"209"},"//images.ctfassets.net/w4dg3cjf42ew/obx1IcApt4s0HxrSr8sSP/ac82bb48e64bb21d756fa8809dec1a56/_BGN0872_2_1_.jpg",{"size":7146603,"image":"283"},"_BGN0872_2[1].jpg","3r3IyiJOlquwikkgMoYuiM","2b6pqr15bygSYem2Og8yGu","9vjuiA89dqoVoNNZmEP7s","46eE1wQL5CQmYmOaIK26cs",{"type":"205","linkType":"32","id":"284"},{"width":445,"height":445},{"width":1723,"height":1803},{"width":600,"height":363},{"width":4750,"height":3167},"1STVrBfdYAMquK226UCOOc"]