BlockGAN is a generative model that learns 3D object-aware scene representations using only unlabelled images

Oct-2-2025, 20:53:18 GMT–Neural Information Processing Systems 

Claims: Although pose is an input to our model, no GT pose labels were used for training.