Questions about training ldm

[pokameng]

Hi @asanakoy @pesser @crowsonkb @theRealSuperMario @to3i I am retraining the LDMs, but the samples are bad. I want to konw, is the number of batch size will affect the sample progress and quality? I use batch size=36 and gpus=3, but you set batch size=48

My config is shown as: model: base_learning_rate: 2.0e-06 target: ldm.models.diffusion.ddpm.LatentDiffusion params: linear_start: 0.0015 linear_end: 0.0195 num_timesteps_cond: 1 log_every_t: 200 timesteps: 1000 first_stage_key: image image_size: 64 channels: 3 monitor: val/loss_simple_ema unet_config: target: ldm.modules.diffusionmodules.openaimodel.UNetModel params: image_size: 64 in_channels: 3 out_channels: 3 model_channels: 224 attention_resolutions:

note: this isn\t actually the resolution but

    # the downsampling factor, i.e. this corresnponds to
    # attention on spatial resolution 8,16,32, as the
    # spatial reolution of the latents is 64 for f4
    - 8
    - 4
    - 2
    num_res_blocks: 2
    channel_mult:
    - 1
    - 2
    - 3
    - 4
    num_head_channels: 32
first_stage_config:
  target: ldm.models.autoencoder.VQModelInterface
  params:
    ckpt_path: /latent-diffusion-main/models/ldm/lsun_beds256/model.ckpt
    embed_dim: 3
    n_embed: 8192
    ddconfig:
      double_z: false
      z_channels: 3
      resolution: 256
      in_channels: 3
      out_ch: 3
      ch: 128
      ch_mult:
      - 1
      - 2
      - 4
      num_res_blocks: 2
      attn_resolutions: []
      dropout: 0.0
    lossconfig:
      target: torch.nn.Identity
cond_stage_config: __is_unconditional__

data: target: main.DataModuleFromConfig params: batch_size: 36 num_workers: 15 wrap: false train: target: ldm.data.lsun.LSUNBedroomsTrain params: size: 256 validation: target: ldm.data.lsun.LSUNBedroomsValidation params: size: 256

lightning: callbacks: image_logger: target: main.ImageLogger params: batch_frequency: 5000 max_images: 8 increase_log_steps: False

trainer: benchmark: True My results are as follow:

[louislee617]

Hi @asanakoy @pesser @crowsonkb @theRealSuperMario @to3i I am retraining the LDMs, but the samples are bad. I want to konw, is the number of batch size will affect the sample progress and quality? I use batch size=36 and gpus=3, but you set batch size=48

My config is shown as: model: base_learning_rate: 2.0e-06 target: ldm.models.diffusion.ddpm.LatentDiffusion params: linear_start: 0.0015 linear_end: 0.0195 num_timesteps_cond: 1 log_every_t: 200 timesteps: 1000 first_stage_key: image image_size: 64 channels: 3 monitor: val/loss_simple_ema unet_config: target: ldm.modules.diffusionmodules.openaimodel.UNetModel params: image_size: 64 in_channels: 3 out_channels: 3 model_channels: 224 attention_resolutions: # note: this isn\t actually the resolution but # the downsampling factor, i.e. this corresnponds to # attention on spatial resolution 8,16,32, as the # spatial reolution of the latents is 64 for f4 - 8 - 4 - 2 num_res_blocks: 2 channel_mult: - 1 - 2 - 3 - 4 num_head_channels: 32 first_stage_config: target: ldm.models.autoencoder.VQModelInterface params: ckpt_path: /latent-diffusion-main/models/ldm/lsun_beds256/model.ckpt embed_dim: 3 n_embed: 8192 ddconfig: double_z: false z_channels: 3 resolution: 256 in_channels: 3 out_ch: 3 ch: 128 ch_mult: - 1 - 2 - 4 num_res_blocks: 2 attn_resolutions: [] dropout: 0.0 lossconfig: target: torch.nn.Identity cond_stage_config: is_unconditional data: target: main.DataModuleFromConfig params: batch_size: 36 num_workers: 15 wrap: false train: target: ldm.data.lsun.LSUNBedroomsTrain params: size: 256 validation: target: ldm.data.lsun.LSUNBedroomsValidation params: size: 256

lightning: callbacks: image_logger: target: main.ImageLogger params: batch_frequency: 5000 max_images: 8 increase_log_steps: False

trainer: benchmark: True My results are as follow:

hello, have you solved your problem about bad reproduced results? And i also account this problem when i trained ldm on CelebAHQ dataset.

[zhenjiaa]

hello,how to obtain these experimental analysis graphs above?

[zyinghua]

hello,how to obtain these experimental analysis graphs above?

In your logs there should be a folder named testtube, and inside (whichever version) there should be a tf folder containing an event file, refer to that event file with command like "tensorboard --logdir=tf", make sure you have tensorboard installed beforehand. Hope this works for you as well :)