4.7 Results: Evaluating LSBD withDLSBD 89 4.7.5 Further Qualitative Results We now show some qualitative results that give more insight into the performance of LSBD-VAE, and how it differs from traditional disentanglement methods. In particular, we investigate the structure of the latent space by showing how data is generated from random latent variables, and from latent space interpolations. (a) COIL-100. (b) ModelNet40. Figure 4.10: Images obtained by decoding latent variables sampled according to the prior over the latent space for different models trained on COIL-100 and ModelNet40. Data generation Inspecting data generated by a model can help understand the structure of the learnt latent space in a qualitative way. Figure 4.10 shows generated data obtained by sampling and decoding ten latent variables for each of the models trained on COIL-100 and ModelNet40. Each latent variable is sampled from the prior over the latent space and decoded to produce an image.
RkJQdWJsaXNoZXIy MjY0ODMw