refactor: consolidate masking logic

imaginairy/modules/diffusion/ddpm.py

@@ -860,11 +860,9 @@ class LatentDiffusion(DDPM):
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
 
     def q_sample(self, x_start, t, noise=None):
-        noise = (
+        if noise is None:
-            noise
+            noise = torch.randn_like(x_start, device="cpu").to(x_start.device)
-            if noise is not None
+
-            else torch.randn_like(x_start, device="cpu").to(x_start.device)
-        )
         return (
             extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
             + extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape)

imaginairy/samplers/base.py

@@ -115,6 +115,30 @@ def get_noise_prediction(
     return noise_pred
 
 
+def mask_blend(noisy_latent, orig_latent, mask, mask_noise, ts, model):
+    """
+    Apply a mask to the noisy_latent.
+
+    ts is a decreasing value between 1000 and 1
+    """
+    assert orig_latent is not None
+    noised_orig_latent = model.q_sample(orig_latent, ts, mask_noise)
+
+    # this helps prevent the weird disjointed images that can happen with masking
+    hint_strength = 0.8
+    # if we're in the first 10% of the steps then don't fully noise the parts
+    # of the image we're not changing so that the algorithm can learn from the context
+    if ts > 900:
+        xdec_orig_with_hints = (
+            noised_orig_latent * (1 - hint_strength) + orig_latent * hint_strength
+        )
+    else:
+        xdec_orig_with_hints = noised_orig_latent
+    noisy_latent = xdec_orig_with_hints * mask + (1.0 - mask) * noisy_latent
+    log_latent(noisy_latent, f"mask-blended noisy_latent {ts}")
+    return noisy_latent
+
+
 def to_torch(x):
     return x.clone().detach().to(torch.float32).to(get_device())
 

imaginairy/samplers/ddim.py

@@ -7,7 +7,7 @@ from tqdm import tqdm
 
 from imaginairy.log_utils import log_latent
 from imaginairy.modules.diffusion.util import extract_into_tensor, noise_like
-from imaginairy.samplers.base import NoiseSchedule, get_noise_prediction
+from imaginairy.samplers.base import NoiseSchedule, get_noise_prediction, mask_blend
 from imaginairy.utils import get_device
 
 logger = logging.getLogger(__name__)
@@ -64,24 +64,25 @@ class DDIMSampler:
         total_steps = timesteps.shape[0]
         noisy_latent = initial_latent
 
+        mask_noise = None
+        if mask is not None:
+            mask_noise = torch.randn_like(noisy_latent, device="cpu").to(
+                noisy_latent.device
+            )
+
         for i, step in enumerate(tqdm(time_range, total=total_steps)):
             index = total_steps - i - 1
             ts = torch.full((batch_size,), step, device=self.device, dtype=torch.long)
 
             if mask is not None:
-                assert orig_latent is not None
+                noisy_latent = mask_blend(
-                xdec_orig = self.model.q_sample(orig_latent, ts)
+                    noisy_latent=noisy_latent,
-                log_latent(xdec_orig, "xdec_orig")
+                    orig_latent=orig_latent,
-                # this helps prevent the weird disjointed images that can happen with masking
+                    mask=mask,
-                hint_strength = 0.8
+                    mask_noise=mask_noise,
-                if i < 2:
+                    ts=ts,
-                    xdec_orig_with_hints = (
+                    model=self.model,
-                        xdec_orig * (1 - hint_strength) + orig_latent * hint_strength
                 )
-                else:
-                    xdec_orig_with_hints = xdec_orig
-                noisy_latent = xdec_orig_with_hints * mask + (1.0 - mask) * noisy_latent
-                log_latent(noisy_latent, "noisy_latent")
 
             noisy_latent, predicted_latent = self.p_sample_ddim(
                 noisy_latent=noisy_latent,

imaginairy/samplers/plms.py

@@ -7,7 +7,7 @@ from tqdm import tqdm
 
 from imaginairy.log_utils import log_latent
 from imaginairy.modules.diffusion.util import extract_into_tensor, noise_like
-from imaginairy.samplers.base import NoiseSchedule, get_noise_prediction
+from imaginairy.samplers.base import NoiseSchedule, get_noise_prediction, mask_blend
 from imaginairy.utils import get_device
 
 logger = logging.getLogger(__name__)
@@ -86,19 +86,14 @@ class PLMSSampler:
             )
 
             if mask is not None:
-                assert orig_latent is not None
+                noisy_latent = mask_blend(
-                xdec_orig = self.model.q_sample(orig_latent, ts, mask_noise)
+                    noisy_latent=noisy_latent,
-                log_latent(xdec_orig, f"xdec_orig i={i} index-{index}")
+                    orig_latent=orig_latent,
-                # this helps prevent the weird disjointed images that can happen with masking
+                    mask=mask,
-                hint_strength = 0.8
+                    mask_noise=mask_noise,
-                if i < 2:
+                    ts=ts,
-                    xdec_orig_with_hints = (
+                    model=self.model,
-                        xdec_orig * (1 - hint_strength) + orig_latent * hint_strength
                 )
-                else:
-                    xdec_orig_with_hints = xdec_orig
-                noisy_latent = xdec_orig_with_hints * mask + (1.0 - mask) * noisy_latent
-                log_latent(noisy_latent, f"x_dec {ts}")
 
             noisy_latent, predicted_latent, noise_pred = self.p_sample_plms(
                 noisy_latent=noisy_latent,