fix repaint

Author: lucidrains

dalle2_pytorch/dalle2_pytorch.py

@@ -516,6 +516,17 @@ def q_sample(self, x_start, t, noise=None):
             extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
         )
 
+    def q_sample_from_to(self, x_from, from_t, to_t, noise = None):
+        shape = x_from.shape
+        noise = default(noise, lambda: torch.randn_like(x_from))
+
+        alpha = extract(self.sqrt_alphas_cumprod, from_t, shape)
+        sigma = extract(self.sqrt_one_minus_alphas_cumprod, from_t, shape)
+        alpha_next = extract(self.sqrt_alphas_cumprod, to_t, shape)
+        sigma_next = extract(self.sqrt_one_minus_alphas_cumprod, to_t, shape)
+
+        return x_from * (alpha_next / alpha) + noise * (sigma_next * alpha - sigma * alpha_next) / alpha
+
     def predict_start_from_noise(self, x_t, t, noise):
         return (
             extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
@@ -2432,14 +2443,18 @@ def p_sample_loop_ddpm(
         is_latent_diffusion = False,
         lowres_noise_level = None,
         inpaint_image = None,
-        inpaint_mask = None
+        inpaint_mask = None,
+        inpaint_resample_times = 5
     ):
         device = self.device
 
         b = shape[0]
         img = torch.randn(shape, device = device)
 
-        if exists(inpaint_image):
+        is_inpaint = exists(inpaint_image)
+        resample_times = inpaint_resample_times if is_inpaint else 1
+
+        if is_inpaint:
             inpaint_image = self.normalize_img(inpaint_image)
             inpaint_image = resize_image_to(inpaint_image, shape[-1], nearest = True)
             inpaint_mask = rearrange(inpaint_mask, 'b h w -> b 1 h w').float()
@@ -2449,31 +2464,40 @@ def p_sample_loop_ddpm(
         if not is_latent_diffusion:
             lowres_cond_img = maybe(self.normalize_img)(lowres_cond_img)
 
-        for i in tqdm(reversed(range(0, noise_scheduler.num_timesteps)), desc = 'sampling loop time step', total = noise_scheduler.num_timesteps):
-            times = torch.full((b,), i, device = device, dtype = torch.long)
-
-            if exists(inpaint_image):
-                # following the repaint paper
-                # https://arxiv.org/abs/2201.09865
-                noised_inpaint_image = noise_scheduler.q_sample(inpaint_image, t = times)
-                img = (img * ~inpaint_mask) + (noised_inpaint_image * inpaint_mask)
-
-            img = self.p_sample(
-                unet,
-                img,
-                times,
-                image_embed = image_embed,
-                text_encodings = text_encodings,
-                cond_scale = cond_scale,
-                lowres_cond_img = lowres_cond_img,
-                lowres_noise_level = lowres_noise_level,
-                predict_x_start = predict_x_start,
-                noise_scheduler = noise_scheduler,
-                learned_variance = learned_variance,
-                clip_denoised = clip_denoised
-            )
+        for time in tqdm(reversed(range(0, noise_scheduler.num_timesteps)), desc = 'sampling loop time step', total = noise_scheduler.num_timesteps):
+            is_last_timestep = time == 0
 
-        if exists(inpaint_image):
+            for r in reversed(range(0, resample_times)):
+                is_last_resample_step = r == 0
+
+                times = torch.full((b,), time, device = device, dtype = torch.long)
+
+                if is_inpaint:
+                    # following the repaint paper
+                    # https://arxiv.org/abs/2201.09865
+                    noised_inpaint_image = noise_scheduler.q_sample(inpaint_image, t = times)
+                    img = (img * ~inpaint_mask) + (noised_inpaint_image * inpaint_mask)
+
+                img = self.p_sample(
+                    unet,
+                    img,
+                    times,
+                    image_embed = image_embed,
+                    text_encodings = text_encodings,
+                    cond_scale = cond_scale,
+                    lowres_cond_img = lowres_cond_img,
+                    lowres_noise_level = lowres_noise_level,
+                    predict_x_start = predict_x_start,
+                    noise_scheduler = noise_scheduler,
+                    learned_variance = learned_variance,
+                    clip_denoised = clip_denoised
+                )
+
+                if is_inpaint and not (is_last_timestep or is_last_resample_step):
+                    # in repaint, you renoise and resample up to 10 times every step
+                    img = noise_scheduler.q_sample_from_to(img, times - 1, times)
+
+        if is_inpaint:
             img = (img * ~inpaint_mask) + (inpaint_image * inpaint_mask)
 
         unnormalize_img = self.unnormalize_img(img)
@@ -2497,7 +2521,8 @@ def p_sample_loop_ddim(
         is_latent_diffusion = False,
         lowres_noise_level = None,
         inpaint_image = None,
-        inpaint_mask = None
+        inpaint_mask = None,
+        inpaint_resample_times = 5
     ):
         batch, device, total_timesteps, alphas, eta = shape[0], self.device, noise_scheduler.num_timesteps, noise_scheduler.alphas_cumprod_prev, self.ddim_sampling_eta
 
@@ -2506,7 +2531,10 @@ def p_sample_loop_ddim(
         times = list(reversed(times.int().tolist()))
         time_pairs = list(zip(times[:-1], times[1:]))
 
-        if exists(inpaint_image):
+        is_inpaint = exists(inpaint_image)
+        resample_times = inpaint_resample_times if is_inpaint else 1
+
+        if is_inpaint:
             inpaint_image = self.normalize_img(inpaint_image)
             inpaint_image = resize_image_to(inpaint_image, shape[-1], nearest = True)
             inpaint_mask = rearrange(inpaint_mask, 'b h w -> b 1 h w').float()
@@ -2519,39 +2547,49 @@ def p_sample_loop_ddim(
             lowres_cond_img = maybe(self.normalize_img)(lowres_cond_img)
 
         for time, time_next in tqdm(time_pairs, desc = 'sampling loop time step'):
-            alpha = alphas[time]
-            alpha_next = alphas[time_next]
+            is_last_timestep = time_next == 0
 
-            time_cond = torch.full((batch,), time, device = device, dtype = torch.long)
+            for r in reversed(range(0, resample_times)):
+                is_last_resample_step = r == 0
 
-            if exists(inpaint_image):
-                # following the repaint paper
-                # https://arxiv.org/abs/2201.09865
-                noised_inpaint_image = noise_scheduler.q_sample(inpaint_image, t = time_cond)
-                img = (img * ~inpaint_mask) + (noised_inpaint_image * inpaint_mask)
+                alpha = alphas[time]
+                alpha_next = alphas[time_next]
 
-            pred = unet.forward_with_cond_scale(img, time_cond, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, lowres_noise_level = lowres_noise_level)
+                time_cond = torch.full((batch,), time, device = device, dtype = torch.long)
 
-            if learned_variance:
-                pred, _ = pred.chunk(2, dim = 1)
+                if is_inpaint:
+                    # following the repaint paper
+                    # https://arxiv.org/abs/2201.09865
+                    noised_inpaint_image = noise_scheduler.q_sample(inpaint_image, t = time_cond)
+                    img = (img * ~inpaint_mask) + (noised_inpaint_image * inpaint_mask)
 
-            if predict_x_start:
-                x_start = pred
-                pred_noise = noise_scheduler.predict_noise_from_start(img, t = time_cond, x0 = pred)
-            else:
-                x_start = noise_scheduler.predict_start_from_noise(img, t = time_cond, noise = pred)
-                pred_noise = pred
+                pred = unet.forward_with_cond_scale(img, time_cond, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, lowres_noise_level = lowres_noise_level)
 
-            if clip_denoised:
-                x_start = self.dynamic_threshold(x_start)
+                if learned_variance:
+                    pred, _ = pred.chunk(2, dim = 1)
 
-            c1 = eta * ((1 - alpha / alpha_next) * (1 - alpha_next) / (1 - alpha)).sqrt()
-            c2 = ((1 - alpha_next) - torch.square(c1)).sqrt()
-            noise = torch.randn_like(img) if time_next > 0 else 0.
+                if predict_x_start:
+                    x_start = pred
+                    pred_noise = noise_scheduler.predict_noise_from_start(img, t = time_cond, x0 = pred)
+                else:
+                    x_start = noise_scheduler.predict_start_from_noise(img, t = time_cond, noise = pred)
+                    pred_noise = pred
+
+                if clip_denoised:
+                    x_start = self.dynamic_threshold(x_start)
+
+                c1 = eta * ((1 - alpha / alpha_next) * (1 - alpha_next) / (1 - alpha)).sqrt()
+                c2 = ((1 - alpha_next) - torch.square(c1)).sqrt()
+                noise = torch.randn_like(img) if not is_last_timestep else 0.
+
+                img = x_start * alpha_next.sqrt() + \
+                      c1 * noise + \
+                      c2 * pred_noise
 
-            img = x_start * alpha_next.sqrt() + \
-                  c1 * noise + \
-                  c2 * pred_noise
+                if is_inpaint and not (is_last_timestep or is_last_resample_step):
+                    # in repaint, you renoise and resample up to 10 times every step
+                    time_next_cond = torch.full((batch,), time_next, device = device, dtype = torch.long)
+                    img = noise_scheduler.q_sample_from_to(img, time_cond, time_next_cond)
 
         if exists(inpaint_image):
             img = (img * ~inpaint_mask) + (inpaint_image * inpaint_mask)
@@ -2658,7 +2696,8 @@ def sample(
         stop_at_unet_number = None,
         distributed = False,
         inpaint_image = None,
-        inpaint_mask = None
+        inpaint_mask = None,
+        inpaint_resample_times = 5
     ):
         assert self.unconditional or exists(image_embed), 'image embed must be present on sampling from decoder unless if trained unconditionally'
 
@@ -2730,7 +2769,8 @@ def sample(
                     noise_scheduler = noise_scheduler,
                     timesteps = sample_timesteps,
                     inpaint_image = inpaint_image,
-                    inpaint_mask = inpaint_mask
+                    inpaint_mask = inpaint_mask,
+                    inpaint_resample_times = inpaint_resample_times
                 )
 
                 img = vae.decode(img)

dalle2_pytorch/version.py

@@ -1 +1 @@
-__version__ = '1.0.3'
+__version__ = '1.0.4'