Closed eeyrw closed 1 year ago
I find some difference between finetune and inference config about item resolution. This is for finetune and resolution is set to 512:
resolution
first_stage_config: target: ldm.models.autoencoder.AutoencoderKL params: embed_dim: 4 monitor: val/rec_loss ddconfig: double_z: true z_channels: 4 resolution: 512 in_channels: 3 out_ch: 3 ch: 128 ch_mult: - 1 - 2 - 4 - 4 num_res_blocks: 2 attn_resolutions: [] dropout: 0.0 lossconfig: target: torch.nn.Identity
This is for inference and resolution is set to 256:
first_stage_config: target: ldm.models.autoencoder.AutoencoderKL params: embed_dim: 4 monitor: val/rec_loss ddconfig: double_z: true z_channels: 4 resolution: 256 in_channels: 3 out_ch: 3 ch: 128 ch_mult: - 1 - 2 - 4 - 4 num_res_blocks: 2 attn_resolutions: [] dropout: 0.0 lossconfig: target: torch.nn.Identity
As far as I know, the first stage model AutoencoderKL is frozen when finetuning. So what's the purpose of changing the resolution? And inference and finetune should share same parameters I think.
I was using Stable Diffusion defaults for the first stage config
I find some difference between finetune and inference config about item
resolution
. This is for finetune and resolution is set to 512:This is for inference and resolution is set to 256:
As far as I know, the first stage model AutoencoderKL is frozen when finetuning. So what's the purpose of changing the
resolution
? And inference and finetune should share same parameters I think.