taming-transformers
taming-transformers copied to clipboard
CompVis
Taming Transformers for High-Resolution Image Synthesis
Hi! Thanks for this wonderful codebase. In the file containing the Encoder and Decoder definitions, I found that `num_res_blocks` was used in the Encoder, and `num_res_blocks + 1` in the...
This pull request contains the following changes: 1. Imports are relative in the `taming` module. 2. Missing `__init__.py` added. We can install it as a plain python package through `setup.py`...
I have trained a transformer model that can generate _256*256_ images, how can I use the **sliding attention window** mechanism mentioned in the paper to generate high resolution images? It...
I try to run the code from latent diffusion model which clones taming-transformer, but I fail as" ModuleNotFoundError: No module named 'cutlit'". cutlit is not a python package, according to...
Fix variable naming errors, embedding_dim and n_embed
Since the transformer take in the quantified image token generated by VQGAN, which codebook has indices (0~n_embed-1), and transformer’s sos token is also set to zero defaultly. Could you tell...
Can anyone help me with this. I am training on custom dataset and i have changes image dataset class as per my need. The error below is coming when i...
Hi, I found the all of the saved reconstructed images are white images under the directory ./logs/ during training(shown below). Is there anyone can tell me the potential reason? 
Hi. There is one output in the decoder class: `print("Working with z of shape {} = {} dimensions.".format(self.z_shape, np.prod(self.z_shape)))` Here the letter z seems to represent the latent vector by...
