Summary

At the beginning of the chapter, we looked at the SynSin model structure, and we gained a deep understanding of the end-to-end process of the model. As mentioned earlier, one interesting approach during the model creation was a differentiable renderer as a part of the training. Also, we saw that the model helps to solve the problem of not having a huge, annotated dataset, or if you don’t have multiple images for test time. That is why this is a state-of-the-art model, which would be easier to use in real-life scenarios. We looked at the pros and cons of the model. Also, we looked at how to initialize the model, train, test, and use new images for inference.

In the next chapter, we will look at the Mesh R-CNN model, which combines two different tasks (object detection and 3D model construction) into one model. We will explore the architecture of the model and test the model performance on a random image.