From 4284c8840de38518694051a46bdf4d9c44eebc18 Mon Sep 17 00:00:00 2001 From: Phil Wang Date: Wed, 8 Jun 2022 16:26:13 -0700 Subject: [PATCH] clipping for continuous time diffusion not working --- .../continuous_time_gaussian_diffusion.py | 12 ++---------- setup.py | 2 +- 2 files changed, 3 insertions(+), 11 deletions(-) diff --git a/denoising_diffusion_pytorch/continuous_time_gaussian_diffusion.py b/denoising_diffusion_pytorch/continuous_time_gaussian_diffusion.py index 1a490c9..addea5b 100644 --- a/denoising_diffusion_pytorch/continuous_time_gaussian_diffusion.py +++ b/denoising_diffusion_pytorch/continuous_time_gaussian_diffusion.py @@ -115,7 +115,6 @@ class ContinuousTimeGaussianDiffusion(nn.Module): loss_type = 'l1', noise_schedule = 'linear', num_sample_steps = 500, - clip_after_noising_during_sampling = False, learned_schedule_net_hidden_dim = 1024, learned_noise_schedule_frac_gradient = 1. # between 0 and 1, determines what percentage of gradients go back, so one can update the learned noise schedule more slowly ): @@ -151,10 +150,6 @@ class ContinuousTimeGaussianDiffusion(nn.Module): self.num_sample_steps = num_sample_steps - # clipping related hyperparameters - - self.clip_after_noising_during_sampling = clip_after_noising_during_sampling - @property def device(self): return next(self.denoise_fn.parameters()).device @@ -216,12 +211,9 @@ class ContinuousTimeGaussianDiffusion(nn.Module): times_next = steps[i + 1] img = self.p_sample(img, times, times_next) - if self.clip_after_noising_during_sampling: - # clip after noise is added. perhaps this is sufficient for Imagen dynamic thresholding? - img.clamp_(-1., 1.) - + img.clamp_(-1., 1.) img = unnormalize_to_zero_to_one(img) - return img.clamp(0., 1.) + return img @torch.no_grad() def sample(self, batch_size = 16): diff --git a/setup.py b/setup.py index c761e85..fd7d128 100644 --- a/setup.py +++ b/setup.py @@ -3,7 +3,7 @@ from setuptools import setup, find_packages setup( name = 'denoising-diffusion-pytorch', packages = find_packages(), - version = '0.17.3', + version = '0.17.4', license='MIT', description = 'Denoising Diffusion Probabilistic Models - Pytorch', author = 'Phil Wang',