On this part, I discuss the manual way to label an object within an image, and getting export xml file as part of Tensorflow object detection.
At that this is a short video or video demonstrating the day when those application data I used or started when I first started last year with machine learning music tends to flow used to forward the labeling process and it was part of the tutorial that I mentioned on the YouTube is one of the steps as far as the labeling process so I thought if you have yes haven’t got the editorial to go head of and look to my YouTube channel and/or my blog and you party find that there but all was you was you taxes or download the application system and a standalone application doesn’t have stuff for us of his daughter think you got to run this in the next together Lindsay Faber so it applies of the same so would you want to do is is you just get familiar with the pulldown menu so here but I’m not in a good tutorial this particular application of this country I might use it so what I’m doing here is all whether these only do is these folders as shown here is a deal with the imaging and then when you actually do the labeling is for you want your annotations of file to be saved and my case year since I’m using tends to flow using the setup has called Flock of format and it’ll save the data in XML file there’s also the he had several people doing labeling you can track the labeling process and and use the verification checkmark here but the mic is on the only one right now the labeling so you don’t bother with that adheres pretty much reading nobody boxes on the image and so forth you can zoom in and out of the images as you need it so well do is an open Open Directory and image directory where have images stored and going there in pretty much gets on the images of started second demonstrating so my case year of an open up this folder here where I’m creating this labeling series video socket haven’t as part of an introduction for the channel and what it does is it’s going to open up all the images in Iraq back here now it easy the first image you see the list of all the images here what I did the doing is end of get one of the videos extracted all the frames that I want to do and and created this image sequence year so the next to you want to do is there is a configuration file that that you can use or update within this program and have predefined labels but in my case year and a user for 314 than a find Cemetery the other one is an identify shadow person and then the other one is a user for my patriots project which is identifying the American flag but what is on sale before forget is that you want to make sure that your application that there’s a checkmark year because as you cycle through the images and you make changes you want to make sure that is automatically saving to that form to the directory she wanted mentioned a checkmark in that you and then you want to us are not open but you want to change the directory where you want your XML files to be safe normally were what I use is I use the same folder where I get image so make is a you same here Cemetery and the selected folder and when the those of us start with the labeling process now when I do my first label you can use these funds here but there’s also this also if you go to the website there is some inputs and will reduce adhere to receive have it available on my maybe the receipt you the led the USA has a list of hotkeys that you can use it every bring it up to his or this report should be easier to test So once cold comes up all open of the with lips and we can download this application so great now this is his though the web webpage we can going here you can download it and it’ll have the full source it’s open source and you go download the your package and if you want to do it for Mac in Berlin makes flavors to tell you how to do it there but anyways I scroll down here and you can see that these hotkeys right here that it be great if it he would’ve put on the application but here you can see that there is hotkeys here:
because the work is expected that he with a lecture so you have these hotkeys here they can use so primary was in and in a use is this study want to create the about box real quick Olmos of in the user to navigate between images real quick then since I have already autosave I won’t musical but I’ll be using this will copy the rectangle label and I be demonstrating how that works so many might you want to tweak the box symbol with around a little bit micro microbrews can use the arrow keys is some shown yourself to to start off when you get deposition here only do is create a about inbox so I create this year and whether things for the Cemetery that I want to the case machine learning is the ability to identify headstones so if I get a video and and some estimates a video and video it is of of Cemetery I wanted to identify headstones to confirm that is a Cemetery right so I let putting here and scope of here just this for these the between purposes instead of putting here is all lowercase one word now this label here is part of a pre-config file and because I’m right for this for this tutorial the 20s for this out to in this intro video I know I create a configuration file specifically for this model for models so I just put labeling to for the purpose of his ambition is to have something so but if he had a predefined config file your labels will be listed here you select that legal but my case and is using of focusing on headstones to click okay so if you have multiple labels for Headstones headstones and I want to capture this a bugs or flies here your labels will start labeling here and they will be color-coded you see this one cyan cyan one other thing that you may want to do is you want to use this display labels see here it shows the labels the name of the labels here so I haven’t any at so oppressed the Internet gives the next image what press and goes back to the next image the previous image right so forward by for a use it is part of a image sequence of the video as the person was moving around with the camera secrecy that headstones morning as a thing that I want to teach machine learning to to learn soul one of the things that that this evidence can do it is it easy to select this rectangle bounding box very good to control the editor copy and to now we can say Want this headstones year and I wanted to do the so you and than Lenexa
and then again press seek a very much copying the label and am teaching machine to learn that Kristi controlled the so wherever your cursor is this the default place so he is going here you just read the label to for how much you wanted to learn I human wanted selectors as well as a defensive watching want the machine to learn so this is kind of blurry to but again this that I’m starting this I don’t know yet how to react when it is training after the training and so forth and again this is the type of video that I’ll be receiving from the the and user son in here I have an extra box with those in the end is inciting have a selected Auxier it doesn’t know to lubricated so you get this is this as well be doing he for the rest of the video select in this headstones think so and night so as of second through you can see the that this highlight here on his images this is going down the best that gives you an indication of where I and he gives you a sense of how much more work you have that did so if you have thousands of images that you need to enable judges sense of where so now this will pretty much is gone cannot recognize it was list of this with he does go back to see how this was here this to start during this year’s strength is a bit still there stated to start learning that it across not quite visible now want to go back and this is a headstones here so here present W select that and it all by default is always for select the last label you facedown see do that you present W again…
Collect this this
again he is creating bounding boxes real quick and if at some point I need to refine my word now is coming reuses application so there’s no community herself make sure that it takes that I sacred to this this is the labeling process for the headstones secrecy that were getting this type of this is that take some time so as part of my project for me I will walk and I will will come data sets with a have an image with the XML formats particular fabric with so we were doing year paranormal paranormal Cemetery CDC this is a image and it’s it’s creating this XML file here is to click that going to open it up this XML file has several times in their one along his path to the file I think this past year is just a general pass but the biggest one is that did was to make sure that it matches this the file name the Jurek getting the width and height and then he gets the name of the first bounding object with the X and Y minimum and exit WiMAX and of course any other bounding boxes that will just progress the show there and Ganley shows you the X and Y minimum
if you have multiple labels on image there are different elect my right not the labels of the same but you can add another label so permissions that is so the another label here headstones headstones that the missile has just this is too far gone so I continue doing this one century ever visiting just the cause barrier is just where the action buried the coffins and that some cement floor under normal wanting to call the headstones to confirm that makes makes things in this is visible is gone to barely make out that even with the human eye and that makes as a matters in so minutes back to reality selected in this is at 30 frames singles 25 frames video and only six selected the first frame out of every 25th grade so I am missing a lot images so in here I will have a lot of data sets for these type of headstones here party have a lot more data sets for different kind headstones and the more I have the morning Jake the mortgage dues the little more accurate the machine learning data string they said will be some that I have reached the after this one so I’ll just move make sure that all the headstones have been labeled as good as what I do now is I would open up my
which is the shadow person and in of a change the same directory to the shadow person to for their now he here because this this one is also in a Cemetery for the captured the show person only do bolt the shadow person and then the cycle through the images the headstones as well what you want to do is enter know who will be doing your labeling or or when you have somebody to label it the best things to do is just to focus on one label name you can do multiple labels an image but after while he taught you start thinking of that Mrs. specific table and get this there’s a lot images and here are elect me 20 images here in more on this list so maybe 40 images solicitor would want to make sure that this is Pascal Wallick and with the start by labeling the dark chattel are no yes I get that it’s right here since here we got a call this shrimpers person a minute myth into dues any space or special characters undisputedly relate Franklin then and cycle through this training use you will start seeing this refine this is this the reason why have more images of this and in its interests good to see what he why I’m doing this is there were two people this is the story this wonderful people in this Cemetery there to YouTube or rare creating their own videos for their you to channels and they all have two independent cameras no they had their it in their own independent cameras and did was captured this shadow figure so the name the name here you can see the images this is person number one point of view image is able down here this way there so many images as the the second person points and this this this image sequence is just a matter of seconds that these make may be like 30 seconds worth of footage at 25 rack frames per second he can imagine this dyslexics hundred Franklin Creek 600 images of this shadow person but for the sake of this labeling is too much time at party make a montage of this video make this video labeling available but was he was the outgrowth so now it does control the existing is better than the highlighting thinking of so sluggish I don’t the incident the congested at select to make sure that it of boxes selected he that it’s right here guys it has and will remember that I’m trying to train this machinery to identify is putting his there’s none here so tractates in… Elect half a body near civic it more visible here in a minute CDC it to the dark and now you see it now spirit there is the this good at this the start insipid dissipating so so is so dark that when the shining the spotlight shines on it is creating its own shadow this this text regulars from the actual video you know and that I did notice as a selected isolate the wrong able Adams don’t see it anymore so go back here Savo Daisy party gave you guys are party screaming at me so is the Sabine is and so it lady ago do is label and select the right one of the bouquet QB she just say thank you doing the same as they will show you of the you different view to there’s none there miracle can control the control he all my runaway he said to see you run the problems I got me like I did the will I make sure that this will not things okay selfishness is that things can get to a point that to see in a vacancy right here so I am going step in the’s you see here this is can run light is from person number two different camera different angle to this event and it’s more fair for camera is on this one this is well within the retraining the machine learning to identify so after this a think she looks away still there so this one person to point of view began returning in this area here right now and should be like halfway to his images so began sending label next so is revered most tedious work right here is doing this label see I know it’s hard to see guys fact that they images of man is good be there and insert of the pixels there’s a difference pixels encoder it’s much darker here than it is from the background just be a analyzed this video this different labeling that would do where a revert the colors and under the be doing that as well topics is easier to see the lease for the human eye my sure of is good be easier for the computer to run from for the be interesting to see the difference in training and then of course you could always it’s just grayscale into this image if he does some kind of fall changes against instead of starting it harder here exited she’s moving out of frame for the scene over the canyon anybody else’s they they realizes saw something in the started getting scared things I guess is the through the so be doing this year this is the labeling processes the has to happen to see once you get into a rhythm he can label a lot of images but after a while here your hands and stay in that position would take some physical and so Franklin F the last so the other thing that I want to do is label this training here and the headstones and then work my way back headstones now a data a lot of people going to do this job here is a tedious but it has to happen guys so when it comes to automations will be looking into how I’m planning to automate this step years will to help us improve our labeling efforts as it is tedious but to get a start somewhere so and you notice part machine learning pleas for the detection this is one of the methods is doing sometimes your to focus will I sent them get enough for headstones here to demonstrate CUC our goal back to cut according you see my shadow person by the boxes a different color than headstones the headstones have assign a color what my shadow boxes recolor and you get the labeling lifetime so you have multiple objects multiple labels lives will be saved and use you see. So that is what I want to do to demos out as is to open up my next project which patriots is this to detect the American flag so the so makes the here that the open of directory change the directory I want to change seven directory to that folder so this is this me driving during the wintertime and that all images will contain the flag sold the be looking at is that this particular flag here on this images sequence the sequence of images is a house has two flights you see so that only do is begin go to the images you start she that the flags are there so write about here is the you see the flags of the putting here to put American the next so it starts learning that the wintertime that’s how the flag link when it’s behind tree branches not this is a – cam I know that the images are not that great but again 17 summits pictures you know the quality of the image that you can receive he can as for the highest quality image but to be honest with you guys some of the guys out there meeting guys been due to doodads man girls been wanting your know the quality of the camera the quality of the forms of be having so we need to include as many variations of images that we can so the straps are so the next one here is coming to McDonald’s American flags for some reason to McDonald’s and had a half staff publisher quite you can barely see it from this distance an image of British serve as so many mightiest Aristarchus you but it’s right so when do this is in the select is partly here just aren’t and taking it from this point me again this is me driving into the office year get some breakfast to coffee so we don’t I working on two different models eyepiece to use the labeling tool is doing label to assess longest of separating the XML files to their respective model this case with you this model online he meant American flag in the labels will be America nothing there’s several variations of the American flag that I remember reading don’t think there is any specific law I’m pretty sure that there is a think but at least for the federal government requires that the flag The 50 states the stripes the 13 stripes and in the colors they’ll represent different something different but there were different variations of the flag and the variations but the sequence of the stars and how in of the appointees of the stars on the points the star supposed to have an effect right now for the majority the flag have five when it starts there were some the had six the some had a centuries for the consumer we get back sequence stars come back after make that turn began we wanted to train to to make sure is the this image here putting out the go now because nowhere like midweight to spring industries party are you for in this flags party. Was that has to known to process your I think of on this time money been that the past presidential George W. Bush no George Bush senior passed away in this party for the flags at half Disrobing here violating days since not signs the rubble stuff here because that remember the I move forward of that truck passes the flag does not come visible in a you is to have highlighted already Ben’s gone back to those less than sold the images that don’t get labels the delegate XML files but as a demonstrate this is this is one of the labeling tools our labeling application tool that you can use to start with all but in my case the agent is another one which is would be a web interface web application which will you can use a web browser to do natural labeling which allows for my subscribers or anybody that want to rigid going join in with the labeling efforts to speed up this process so anyways I hope this was helpful to you guys again our goal is you went to the tutorial through the tutorial he uses this label applications such for a demonstrated easy it’s easy to use you can use this application for the next and a has Macintoshes go to websites composites ~have compiler for those operating systems and then was you creates labeling labels to create matching XML file that concedes that they done than you continue want to tutorial try to massage the training the training data to start the training process with intense I nothing is said for me right now thank you hopefully this is a video several to yes