Image Annotation Lab Tutorial

hello everyone today i will show you how you can label objects in images and videos with image annotation lab the first thing that we must do is to create a new project or to open an existing project let's first create a new project by selecting a collection of images on the hard drive so we click on file create project and then select the folder where our images reside the images of the new projects are now visible in the browser view on the left side of the main window if we click on one of these images we make this image the active image this is now displayed in the central window the next step is to create a set of names for the annotation regions we want to draw these names are called labels to do so we click on the button edit labels and add a couple of new labels let's add a label called person another label called animal one more label called mushroom another label called background for the background region and finally we also delete the empty label which is always present as a starter label by clicking on edit selecting delete and removing it okay so we now see our four labels that we have created animal background mushroom and person we are now all set to create our first annotations we start by adding a manual annotation region using the polygon tool so let's select that tool from the tools available on the left side of the main image viewer we want to add a region called mushroom so let's also select the right label from our set of available labels in the drop down box and change the color for this label to blue now we can start labeling our first region by doing a couple of left clicks along the contour of the object of interest once we have finished we close the region with a double click if we make a mistake we can easily correct it by adding something to the region outlined so far or by subtracting something from it let us correct the missing part of our mushroom whether we add something or subtract something is determined by the plus or minus buttons located at the lower left hand side of the image viewer so in order to subtract the newly drawn region from the existing annotation you either click on the minus button to activate it or you press the s key to only temporarily activate the subtract mode as long as you press that key so let us try that and make a hole in our nice mushroom object well now that we have created our first annotation let's see how we can delete it again in order to delete a labeled region we first have to select it this is done by making a right click somewhere on the colored area of the annotation this will outline the region with a thick yellow contour once the region is selected we can delete it by simply pressing the delete key don't worry if you delete something by mistake you can always undo this by pressing ctrl z imageannotationlab allows annotation regions with different labels to interact with each other interaction means that you can add subtract or intersect such regions this makes the labeling of complex scenes much easier assume we want to label everything but the mushroom on this image is background all we have to do is to first select background as our new next label and then draw a rectangular annotation on the whole image we now have the background region overlap the mushroom region in order to subtract the mushroom from the background we simply select these two areas and press the ctrl plus minus key this will separate the mushroom nicely from the background note that the hole in the mushroom is now also marked as background we can correct that easily by first deselecting everything by a right click outside of the image and then selecting the contour of the hole and the inner background area and deleting them the mushroom is now complete again without a hole up to now we have drawn our annotation regions by hand this is fine as long as the regions to label are not too complex however once you would like to have pixel precise labeling this becomes very time consuming and tedious fortunately image annotation lab contains powerful artificial intelligence tools to assist you one of these tools is called interactive segmentation interactive segmentation is activated by clicking on the ai button next to the image viewer once the ai button is activated you can draw a rectangle around the object you would like to label to demonstrate this let us first choose a different label we select the label person and we also select a different image now we draw a rectangular region around the person this opens a dialogue for interactive segmentation with interactive segmentation we can add positive and negative clicks to an image and thus guide the annotation process a positive click which is simply a left click will ask the software to add the region around the click to the labeled region and a negative click which is shift plus a left click will ask interactive segmentation to remove the area from the labeled region the tool is smart enough to have knowledge about what constitutes an object so that the labeling process only requires very few clicks to finish once we have finished with interactive segmentation the labeled region is transferred back to the main window and assigned the currently active label another powerful feature of image annotation lab is its ability to process video files and to annotate sequences of video frames automatically to show this let us create a new project however this time we do not use images stored on the hard drive but will extract a sequence of frames from a video file so we select file and then create project from video file we now choose our video file and the location where we would like to store the extracted frames the video extraction dialog will show you important information about the chosen video file like the width and height of each frame as well as the frame rate and the overall number of frames in the video you can also define how you would like to extract frames from the video for example if you would like to extract every frame or only every 10th or 20th and how you would like to store the extracted frames let us extract every 10th frame and reduce the width of each frame to a more manageable size of 1100 pixels the height of the frame will be adjusted accordingly to keep the original aspect ratio after clicking ok the frames from the video will be extracted and a new project is created the new project consists of a sequence of images showing a tiger slowly walking from left to right our aim is to label the tiger by precisely drawing its outlines on every frame in total we have 41 frames so labeling each image individually is quite an effort luckily we can resort to another smart tool in image annotation lab which is called video object segmentation with video object segmentation we only have to label one or a few frames of a sequence and the computer will do the rest of the work first we select a new active label we now choose animal then we need to label the tiger on the first frame we can do that manually or much better use interactive segmentation once again now we must tell video object segmentation about the objects on the start frame therefore we select the tiger and choose tools video object segmentation and set vos initial frame once we have done that we can try to label the rest of the frames automatically so let us select tools video object segmentation and vos on image range in the upcoming dialog we now select all images of the tiger that we want the computer to annotate for us so we left click on the second frame and shift plus left click on the last frame of our sequence depending on your computer and especially if you have a cuda enabled graphics card the processing of each frame will take something between 50 milliseconds and a couple of seconds as we can see the results are very convincing the tiger was nicely segmented in every frame and this in a fraction of the time it would take us to do this manually in case some of the segmentations are not satisfactory you can correct them manually and add the corrected frames as additional reference frames to video object segmentation this will improve the result on the next run of the tool this brings us to the end of this video presentation there are many more features in image annotation lab that we have not covered i invite you to try it for yourself and download it for free happy labeling

Channel: 4SmartMachines

Views: 12,046

Keywords: training data, machine learning, video labeling, video annotation, image labeling, image annotation, deep learning, image annotation tool, image labeling tool, image annotation software, image labeling software, video annotation tool, video labeling software, video labeling tool

Id: scL1oTH_U_0

Length: 9min 35sec (575 seconds)

Published: Wed Feb 03 2021