Image Labeling API | Automatically Label Computer Vision Data

Key Takeaways

hi everybody it's Peter from roboflow here let me show you how easy it is to use ruleflow inference API to how to annotate your images this is really useful in at least two cases first of all when you already have a data set and when you already trained their model using roboflow but you notice that the model under performs in some specific edge cases it is very common practice to gather new images that would represent that edge case and add them to your data set and retrain the model you can leverage your existing model and roboflow inference API to Auto annotate those images second of all when you would like to Kickstart your new data set and you noticed some model in roboflow universe that have some classes that you would like to use or some very similar objects then you could leverage that model to Auto annotate your new images okay so here I am logged into my roboflow account and I would like to add new images into my food herbal player detection project I already have the data set that is annotated and I already have the model that is trained using the data set I just want to add new images to do that we can use roboflow Auto annotate script that is accessible in one of our repositories the link to that script is in the description okay so let's follow installation instructions first of all we need to clone the Repository and then we need to get inside the directory containing the script the next step is to create python virtual environment to separate your code over here from other projects that you have on your machine so let's copy that command to create vmf let's get into that VM now we need to install the script inside our virtual environment and we are basically ready to go the next step is to set up virtual environment variable with your roboflow API key so let's copy that part of the command into the terminal and the API key can be found in your roboflow profile under the settings roboflow roboflow API and the private key over here is the one that we are looking for so let's copy that go back to the terminal and instead of those three dots let's paste the API key over here and hit enter it is important to keep your API key as private as possible don't commit that into your repository especially when it's public now the last step is to actually execute the command to annotate so let's copy that from the repository readme into the terminal and we need to fill those missing pieces over here in the meantime I decided to increase the font size in the terminal a bit so it would be a little bit easier for you to read what I'm doing the first thing is input image directory I already prepared a set of images that I would like to annotate and they are located over here in that directory so I will use PWD to have full path to that directory and inside that directory when I hit LS I see images directory so here under source image directory I will just paste the path to the general directory and then I will type images at the end I want the results to be safe in the same directory just under different sub directory so let's call it labels now I need to provide information about the model that I would like to use to Auto annotate so let's go back to roboflow into my projects football player detection deploy and over here I need to get project ID and model version so I have my project ID in the path over here so it's football player detection and those four characters at the end so let's copy that and the version of the model that I would like to use is the second one and the only thing that is left to do is press the enter and let it annotate automatically so you immediately see that we see roboflow workspace being fetched and roboflow project being fed and now we just Loop over the images and Save The annotation Json in the labels directory we can see it happening in the real time if we go to labels we see that the labels for already processed images already here and it's done so let's test if my newly annotated images will get correctly uploaded into my already existing project I rearranged my windows a little bit so it would be easy for me to drag and drop images and labels so let's now go to annotate upload more images grab those two directories and just drop them over here and we immediately see that the images were loaded but also on the thumbnail we see that annotations are also visible you can examine That Into The annotation tool and here on the slightly larger screen we see that all the classes bottle keeper player referee were loaded and correctly annotated now we are ready for data set refined so over here for example our current model we're not able to correctly detect that player so we would be able to just manually add that bounding box around the player over here and continue the work for all other images this way we can add more images but most of the heavy lifting is already done by our model and that's it I hope that you find it useful I definitely do I use it all the time during model refinement to spend as little time as possible on labeling