The three expert reviewers agree on the merits of the submission, as it introduces and validates an interesting and elegant approach that addresses an important and challenging problem of learning deformable models from data. Based on the feedback from the reviewers, the acceptance is recommended, potentially as a spotlight. A side note: the authors rightfully draw connection of their approach to the auto-decoder model [39]. The authors of [39], however, missed an earlier work of Bojanowski et al. ICML18 (Optimizing the Latent Space of Generative Networks) that had previously introduced and popularized "encoder-free models". This lead to unfortunate fragmentation/duplication of terminology. The authors are encouraged to consider the connection to and the terminology from Bojanowski et al. when discussing "auto-decoders"/"encoder-free models" in the final version.