Imaginghub Forum - 3D modeling for image augmentation

The problem of image data scarcity is a serious issue on the way of computer vision models improvement. Especially it concerns the very narrow areas, where you should, for example, classify some objects which images cannot be found on the Internet. Data augmentation can be a solution to this problem. My question is whether the 3D models of such objects can help in data augmentation and if yes, then what is the general process of performing data augmentation with the help of 3d modeling?

Answers

During the computer vision models training the more data you have the better. The best models can recognize objects from different angles and from different sides. If you need to work with some unique objects and you really want to achieve good results, then 3d modeling can help you. You should create a 3 model of the object and take a lot of pictures of the object from different sides, angles, etc. You can make small variations in the object, try to place the object behind other objects. In other words, with 3d modeling, you can simulate many real-world situations where your computer vision model should work well.

Thanks. Are there some downsides of this approach?

Surely, as well as with many other approaches. In my opinion, the major disadvantages are cost and time. Imagine how much time and cost are needed to create a 3d model and then to play with it. You should know how to work with the data modeling tools or hire someone who can do this. But if you don't have any other sources of data and you need to train the model well in any means, then this approach definitely should be tried.

Add Answer

3D modeling for image augmentation

3D modeling for image augmentation

Answers

Need support?