2021-06-03 03:47:32 +00:00
|
|
|
import torch
|
|
|
|
|
|
|
|
from models.diffusion.gaussian_diffusion import GaussianDiffusion, get_named_beta_schedule
|
2021-06-05 19:40:32 +00:00
|
|
|
from models.diffusion.resample import create_named_schedule_sampler, LossAwareSampler
|
2021-06-04 23:13:16 +00:00
|
|
|
from models.diffusion.respace import space_timesteps, SpacedDiffusion
|
2021-06-03 03:47:32 +00:00
|
|
|
from trainer.inject import Injector
|
|
|
|
from utils.util import opt_get
|
|
|
|
|
|
|
|
|
|
|
|
# Injects a gaussian diffusion loss as described by OpenAIs "Improved Denoising Diffusion Probabilistic Models" paper.
|
|
|
|
# Largely uses OpenAI's own code to do so (all code from models.diffusion.*)
|
|
|
|
class GaussianDiffusionInjector(Injector):
|
|
|
|
def __init__(self, opt, env):
|
|
|
|
super().__init__(opt, env)
|
|
|
|
self.generator = opt['generator']
|
2021-06-04 23:13:16 +00:00
|
|
|
self.output_variational_bounds_key = opt['out_key_vb_loss']
|
|
|
|
self.output_x_start_key = opt['out_key_x_start']
|
2021-06-03 03:47:32 +00:00
|
|
|
opt['diffusion_args']['betas'] = get_named_beta_schedule(**opt['beta_schedule'])
|
2021-06-06 22:35:37 +00:00
|
|
|
opt['diffusion_args']['use_timesteps'] = space_timesteps(opt['beta_schedule']['num_diffusion_timesteps'],
|
|
|
|
[opt['beta_schedule']['num_diffusion_timesteps']])
|
2021-06-04 23:13:16 +00:00
|
|
|
self.diffusion = SpacedDiffusion(**opt['diffusion_args'])
|
2021-06-03 03:47:32 +00:00
|
|
|
self.schedule_sampler = create_named_schedule_sampler(opt['sampler_type'], self.diffusion)
|
|
|
|
self.model_input_keys = opt_get(opt, ['model_input_keys'], [])
|
|
|
|
|
|
|
|
def forward(self, state):
|
|
|
|
gen = self.env['generators'][self.opt['generator']]
|
|
|
|
hq = state[self.input]
|
|
|
|
model_inputs = {k: state[v] for k, v in self.model_input_keys.items()}
|
|
|
|
t, weights = self.schedule_sampler.sample(hq.shape[0], hq.device)
|
2021-06-04 23:13:16 +00:00
|
|
|
diffusion_outputs = self.diffusion.training_losses(gen, hq, t, model_kwargs=model_inputs)
|
2021-06-05 19:40:32 +00:00
|
|
|
if isinstance(self.schedule_sampler, LossAwareSampler):
|
|
|
|
self.schedule_sampler.update_with_local_losses(t, diffusion_outputs['losses'])
|
2021-06-04 23:13:16 +00:00
|
|
|
return {self.output: diffusion_outputs['mse'],
|
|
|
|
self.output_variational_bounds_key: diffusion_outputs['vb'],
|
|
|
|
self.output_x_start_key: diffusion_outputs['x_start_predicted']}
|
2021-06-03 03:47:32 +00:00
|
|
|
|
|
|
|
|
2021-07-26 22:27:31 +00:00
|
|
|
class AutoregressiveGaussianDiffusionInjector(Injector):
|
|
|
|
def __init__(self, opt, env):
|
|
|
|
super().__init__(opt, env)
|
|
|
|
self.generator = opt['generator']
|
|
|
|
self.output_variational_bounds_key = opt['out_key_vb_loss']
|
|
|
|
self.output_x_start_key = opt['out_key_x_start']
|
|
|
|
opt['diffusion_args']['betas'] = get_named_beta_schedule(**opt['beta_schedule'])
|
|
|
|
opt['diffusion_args']['use_timesteps'] = space_timesteps(opt['beta_schedule']['num_diffusion_timesteps'],
|
|
|
|
[opt['beta_schedule']['num_diffusion_timesteps']])
|
|
|
|
self.diffusion = SpacedDiffusion(**opt['diffusion_args'])
|
|
|
|
self.schedule_sampler = create_named_schedule_sampler(opt['sampler_type'], self.diffusion)
|
|
|
|
self.model_input_keys = opt_get(opt, ['model_input_keys'], [])
|
|
|
|
self.model_output_keys = opt['model_output_keys']
|
|
|
|
self.model_eps_pred_key = opt['prediction_key']
|
|
|
|
|
|
|
|
def forward(self, state):
|
|
|
|
gen = self.env['generators'][self.opt['generator']]
|
|
|
|
hq = state[self.input]
|
|
|
|
model_inputs = {k: state[v] for k, v in self.model_input_keys.items()}
|
|
|
|
t, weights = self.schedule_sampler.sample(hq.shape[0], hq.device)
|
|
|
|
diffusion_outputs = self.diffusion.autoregressive_training_losses(gen, hq, t, self.model_output_keys,
|
|
|
|
self.model_eps_pred_key,
|
|
|
|
model_kwargs=model_inputs)
|
|
|
|
if isinstance(self.schedule_sampler, LossAwareSampler):
|
|
|
|
self.schedule_sampler.update_with_local_losses(t, diffusion_outputs['losses'])
|
|
|
|
outputs = {k: diffusion_outputs[k] for k in self.model_output_keys}
|
|
|
|
outputs.update({self.output: diffusion_outputs['mse'],
|
|
|
|
self.output_variational_bounds_key: diffusion_outputs['vb'],
|
|
|
|
self.output_x_start_key: diffusion_outputs['x_start_predicted']})
|
|
|
|
return outputs
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-03 03:47:32 +00:00
|
|
|
# Performs inference using a network trained to predict a reverse diffusion process, which nets a image.
|
|
|
|
class GaussianDiffusionInferenceInjector(Injector):
|
|
|
|
def __init__(self, opt, env):
|
|
|
|
super().__init__(opt, env)
|
2021-06-16 22:26:36 +00:00
|
|
|
use_ddim = opt_get(opt, ['use_ddim'], False)
|
2021-06-03 03:47:32 +00:00
|
|
|
self.generator = opt['generator']
|
2021-06-11 21:31:10 +00:00
|
|
|
self.output_batch_size = opt['output_batch_size']
|
|
|
|
self.output_scale_factor = opt['output_scale_factor']
|
|
|
|
self.undo_n1_to_1 = opt_get(opt, ['undo_n1_to_1'], False) # Explanation: when specified, will shift the output of this injector from [-1,1] to [0,1]
|
2021-06-03 03:47:32 +00:00
|
|
|
opt['diffusion_args']['betas'] = get_named_beta_schedule(**opt['beta_schedule'])
|
2021-06-16 22:26:36 +00:00
|
|
|
if use_ddim:
|
|
|
|
spacing = "ddim" + str(opt['respaced_timestep_spacing'])
|
|
|
|
else:
|
|
|
|
spacing = [opt_get(opt, ['respaced_timestep_spacing'], opt['beta_schedule']['num_diffusion_timesteps'])]
|
|
|
|
opt['diffusion_args']['use_timesteps'] = space_timesteps(opt['beta_schedule']['num_diffusion_timesteps'], spacing)
|
2021-06-04 23:13:16 +00:00
|
|
|
self.diffusion = SpacedDiffusion(**opt['diffusion_args'])
|
2021-06-16 22:26:36 +00:00
|
|
|
self.sampling_fn = self.diffusion.ddim_sample_loop if use_ddim else self.diffusion.p_sample_loop
|
2021-06-03 03:47:32 +00:00
|
|
|
self.model_input_keys = opt_get(opt, ['model_input_keys'], [])
|
2021-06-14 15:14:30 +00:00
|
|
|
self.use_ema_model = opt_get(opt, ['use_ema'], False)
|
2021-06-21 16:38:07 +00:00
|
|
|
self.noise_style = opt_get(opt, ['noise_type'], 'random') # 'zero', 'fixed' or 'random'
|
2021-06-03 03:47:32 +00:00
|
|
|
|
|
|
|
def forward(self, state):
|
2021-06-14 15:14:30 +00:00
|
|
|
if self.use_ema_model:
|
|
|
|
gen = self.env['emas'][self.opt['generator']]
|
|
|
|
else:
|
|
|
|
gen = self.env['generators'][self.opt['generator']]
|
2021-06-11 21:31:10 +00:00
|
|
|
model_inputs = {k: state[v][:self.output_batch_size] for k, v in self.model_input_keys.items()}
|
2021-06-03 03:47:32 +00:00
|
|
|
gen.eval()
|
|
|
|
with torch.no_grad():
|
2021-06-11 21:31:10 +00:00
|
|
|
output_shape = (self.output_batch_size, 3, model_inputs['low_res'].shape[-2] * self.output_scale_factor,
|
|
|
|
model_inputs['low_res'].shape[-1] * self.output_scale_factor)
|
2021-06-21 16:38:07 +00:00
|
|
|
noise = None
|
|
|
|
if self.noise_style == 'zero':
|
|
|
|
noise = torch.zeros(output_shape, device=model_inputs['low_res'].device)
|
|
|
|
elif self.noise_style == 'fixed':
|
|
|
|
if not hasattr(self, 'fixed_noise') or self.fixed_noise.shape != output_shape:
|
|
|
|
self.fixed_noise = torch.randn(output_shape, device=model_inputs['low_res'].device)
|
|
|
|
noise = self.fixed_noise
|
2021-06-16 22:26:36 +00:00
|
|
|
gen = self.sampling_fn(gen, output_shape, noise=noise, model_kwargs=model_inputs, progress=True)
|
2021-06-11 21:31:10 +00:00
|
|
|
if self.undo_n1_to_1:
|
|
|
|
gen = (gen + 1) / 2
|
2021-06-03 03:47:32 +00:00
|
|
|
return {self.output: gen}
|