make ldm work, add classifier free guidence

models/vision/latent_diffusion/modeling_latent_diffusion.py

@@ -2,10 +2,10 @@
 import math
 
 import numpy as np
+import tqdm
 import torch
 import torch.nn as nn
 
-import tqdm
 from diffusers import DiffusionPipeline
 from diffusers.configuration_utils import ConfigMixin
 from diffusers.modeling_utils import ModelMixin
@@ -740,30 +740,29 @@ class DiagonalGaussianDistribution(object):
 
     def kl(self, other=None):
         if self.deterministic:
-            return torch.Tensor([0.0])
+            return torch.Tensor([0.])
         else:
             if other is None:
-                return 0.5 * torch.sum(torch.pow(self.mean, 2) + self.var - 1.0 - self.logvar, dim=[1, 2, 3])
+                return 0.5 * torch.sum(torch.pow(self.mean, 2)
+                                       + self.var - 1.0 - self.logvar,
+                                       dim=[1, 2, 3])
             else:
                 return 0.5 * torch.sum(
                     torch.pow(self.mean - other.mean, 2) / other.var
-                    + self.var / other.var
-                    - 1.0
-                    - self.logvar
-                    + other.logvar,
-                    dim=[1, 2, 3],
-                )
+                    + self.var / other.var - 1.0 - self.logvar + other.logvar,
+                    dim=[1, 2, 3])
 
-    def nll(self, sample, dims=[1, 2, 3]):
+    def nll(self, sample, dims=[1,2,3]):
         if self.deterministic:
-            return torch.Tensor([0.0])
+            return torch.Tensor([0.])
         logtwopi = np.log(2.0 * np.pi)
-        return 0.5 * torch.sum(logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var, dim=dims)
+        return 0.5 * torch.sum(
+            logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var,
+            dim=dims)
 
     def mode(self):
         return self.mean
 
-
 class AutoencoderKL(ModelMixin, ConfigMixin):
     def __init__(
         self,
@@ -835,7 +834,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin):
             give_pre_end=give_pre_end,
         )
 
-        self.quant_conv = torch.nn.Conv2d(2 * z_channels, 2 * embed_dim, 1)
+        self.quant_conv = torch.nn.Conv2d(2*z_channels, 2*embed_dim, 1)
         self.post_quant_conv = torch.nn.Conv2d(embed_dim, z_channels, 1)
 
     def encode(self, x):
@@ -864,7 +863,7 @@ class LatentDiffusion(DiffusionPipeline):
         super().__init__()
         self.register_modules(vqvae=vqvae, bert=bert, tokenizer=tokenizer, unet=unet, noise_scheduler=noise_scheduler)
 
-    def __call__(self, prompt, batch_size=1, generator=None, torch_device=None, eta=0.0, num_inference_steps=50):
+    def __call__(self, prompt, batch_size=1, generator=None, torch_device=None, eta=0.0, guidance_scale=1.0, num_inference_steps=50):
         # eta corresponds to η in paper and should be between [0, 1]
 
         if torch_device is None:
@@ -874,6 +873,10 @@ class LatentDiffusion(DiffusionPipeline):
         self.vqvae.to(torch_device)
         self.bert.to(torch_device)
         
+        if guidance_scale != 1.0:
+            uncond_input = self.tokenizer([""], padding="max_length", max_length=77, return_tensors='pt').to(torch_device)
+            uncond_embeddings = self.bert(uncond_input.input_ids)[0]
+        
         # get text embedding
         text_input = self.tokenizer(prompt, padding="max_length", max_length=77, return_tensors='pt').to(torch_device)
         text_embedding = self.bert(text_input.input_ids)[0]
@@ -886,45 +889,74 @@ class LatentDiffusion(DiffusionPipeline):
             device=torch_device,
             generator=generator,
         )
+        
+        # See formulas (12) and (16) of DDIM paper https://arxiv.org/pdf/2010.02502.pdf
+        # Ideally, read DDIM paper in-detail understanding
+
+        # Notation (<variable name> -> <name in paper>
+        # - pred_noise_t -> e_theta(x_t, t)
+        # - pred_original_image -> f_theta(x_t, t) or x_0
+        # - std_dev_t -> sigma_t
+        # - eta -> η
+        # - pred_image_direction -> "direction pointingc to x_t"
+        # - pred_prev_image -> "x_t-1"
         for t in tqdm.tqdm(reversed(range(num_inference_steps)), total=num_inference_steps):
-            # get actual t and t-1
+            # 1. predict noise residual
+            if guidance_scale == 1.0:
+                timesteps = torch.tensor([inference_step_times[t]] * image.shape[0], device=torch_device)
+                context = text_embedding
+                image_in = image
+            else:
+                image_in = torch.cat([image] * 2)
+                timesteps = torch.tensor([inference_step_times[t]] * image.shape[0], device=torch_device)
+                context = torch.cat([uncond_embeddings, text_embedding])
+            
+            with torch.no_grad():
+                pred_noise_t = self.unet(image_in, timesteps, context=context)
+            
+            if guidance_scale != 1.0:
+                pred_noise_t_uncond, pred_noise_t = pred_noise_t.chunk(2)
+                pred_noise_t = pred_noise_t_uncond + guidance_scale * (pred_noise_t - pred_noise_t_uncond)
+                    
+            # 2. get actual t and t-1
             train_step = inference_step_times[t]
             prev_train_step = inference_step_times[t - 1] if t > 0 else -1
 
-            # compute alphas
+            # 3. compute alphas, betas
             alpha_prod_t = self.noise_scheduler.get_alpha_prod(train_step)
             alpha_prod_t_prev = self.noise_scheduler.get_alpha_prod(prev_train_step)
-            alpha_prod_t_rsqrt = 1 / alpha_prod_t.sqrt()
-            alpha_prod_t_prev_rsqrt = 1 / alpha_prod_t_prev.sqrt()
-            beta_prod_t_sqrt = (1 - alpha_prod_t).sqrt()
-            beta_prod_t_prev_sqrt = (1 - alpha_prod_t_prev).sqrt()
-
-            # compute relevant coefficients
-            coeff_1 = (
-                (alpha_prod_t_prev - alpha_prod_t).sqrt()
-                * alpha_prod_t_prev_rsqrt
-                * beta_prod_t_prev_sqrt
-                / beta_prod_t_sqrt
-                * eta
-            )
-            coeff_2 = ((1 - alpha_prod_t_prev) - coeff_1**2).sqrt()
+            beta_prod_t = 1 - alpha_prod_t
+            beta_prod_t_prev = 1 - alpha_prod_t_prev
 
-            # model forward
-            with torch.no_grad():
-                train_step = torch.tensor([train_step] * image.shape[0], device=torch_device)
-                noise_residual = self.unet(image, train_step, context=text_embedding)
+            # 4. Compute predicted previous image from predicted noise
+            # First: compute predicted original image from predicted noise also called
+            # "predicted x_0" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_original_image = (image - beta_prod_t.sqrt() * pred_noise_t) / alpha_prod_t.sqrt()
 
-            # predict mean of prev image
-            pred_mean = alpha_prod_t_rsqrt * (image - beta_prod_t_sqrt * noise_residual)
-            pred_mean = torch.clamp(pred_mean, -1, 1)
-            pred_mean = (1 / alpha_prod_t_prev_rsqrt) * pred_mean + coeff_2 * noise_residual
+            # Second: Clip "predicted x_0"
+#             pred_original_image = torch.clamp(pred_original_image, -1, 1)
 
-            # if eta > 0.0 add noise. Note eta = 1.0 essentially corresponds to DDPM
+            # Third: Compute variance: "sigma_t(η)" -> see formula (16)
+            # σ_t = sqrt((1 − α_t−1)/(1 − α_t)) * sqrt(1 − α_t/α_t−1)
+            std_dev_t = (beta_prod_t_prev / beta_prod_t).sqrt() * (1 - alpha_prod_t / alpha_prod_t_prev).sqrt()
+            std_dev_t = eta * std_dev_t
+
+            # Fourth: Compute "direction pointing to x_t" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_image_direction = (1 - alpha_prod_t_prev - std_dev_t**2).sqrt() * pred_noise_t
+
+            # Fifth: Compute x_t without "random noise" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_prev_image = alpha_prod_t_prev.sqrt() * pred_original_image + pred_image_direction
+
+            # 5. Sample x_t-1 image optionally if η > 0.0 by adding noise to pred_prev_image
+            # Note: eta = 1.0 essentially corresponds to DDPM
             if eta > 0.0:
                 noise = self.noise_scheduler.sample_noise(image.shape, device=image.device, generator=generator)
-                image = pred_mean + coeff_1 * noise
+                prev_image = pred_prev_image + std_dev_t * noise
             else:
-                image = pred_mean
+                prev_image = pred_prev_image
+
+            # 6. Set current image to prev_image: x_t -> x_t-1
+            image = prev_image
 
         image = 1 /  0.18215 * image
         image = self.vqvae.decode(image)