Jun. 8, 2021
Synthetic data can now radically accelerate the creation of AI models for computer vision. Watch this 5 minute vision to see how it works. Then, to learn more and to sign up for a demo of the Chooch AI 3D Synthetic Data Generation tool on the computer vision platform, please visit Synthetic Data Generation
Here is a transcript of the video.
Hello. So right now you’re on datasets section on Chooch website. When you click on my object datasets, you’re headed toward datasets. So I have created an empty 3D synthetic test folder, a dataset. So 3D synthetic data allows you to generate synthetic images by just using one CAD model. So to do this on Chooch we press on synthetic data, and then we upload files from our local. So right now the system allows OBJ, MTL, and STL files. So we upload these, and once the upload is finished you can see the files here. So when you click on it, the model is loaded in the viewer. In the viewer, you can check if the model is right, you can just play with it and so on. And on the left bottom, you can see the other files that you have uploaded with this model.
So you just press generate synthetic data, with the slider, you can change how many images you would like to generate, right now I’ll just stick to 10. In advanced settings you can play with the resolution of the images that you would like to generate. We can also generate images in gray scale as an option. In object and whole object in annotation mode refers to annotating the whole object as a whole or parts of the object. So if the groups are predefined in the object file, you can select this option. So that for example, it can generate multiple bounding boxes for one object if the object has subgroups in it. And this mode refers to scale can choose close, far or very far.
For example, if you would like to detect something very far, it’s better to generate images that the object is far away. So for now we just select all and you save it. And if you’d like to use one of your themes, you can just say my themes and from our datasets or from your raw data, you can upload your own themes. For now, I’ll just choose or to custom choose teams since this is a plane and I’ll just choose sky and press generate. So the job has started in the background. We can just exit from this manning.
On the left, in the view job section, you can see something’s going on, the top row is your latest job that you had triggered. The previous ones, as you can see is sort of job completed. As you can see this one, the current status is of this one is in process. Also for API users you’re getting a unique ID of the job and also you’re getting the date that you have triggered the job. So, okay. The job is done so we can see it’s completed. What we have to do is just refresh the page and you’ll see our plane is being generated, print plane images. So let’s check a couple of them, let’s check this one. So this is, as you can see the annotations, the bounding boxes already here, and you just put your cursor on it and it says 3D model. There are different ones to scale like that one.
This is for example, a very far plane. So you can generate any amount of data. Let’s check this one. This is from one image from bottom view. So if you like, you can, the classes are listed here. You can rename them and let’s rename it to ATR 42, which is the plane’s actual model. So after this is updated, then you go and press on the images and then you put your cursor and you can see the name has changed. So 3D synthetic data allows you to generate infinite amount of images. And this is useful for coming up with a bit better models. Thank you.