feat: added cutout augs to vqgan-clip

app.py

@@ -44,6 +44,7 @@ def generate_image(
     mse_weight_decay: float = 0,
     mse_weight_decay_steps: int = 0,
     tv_loss_weight: float = 1e-3,
+    use_cutout_augmentations: bool = True,
 ) -> None:
 
     ### Init -------------------------------------------------------------------
@@ -61,6 +62,7 @@ def generate_image(
         mse_weight_decay=mse_weight_decay,
         mse_weight_decay_steps=mse_weight_decay_steps,
         tv_loss_weight=tv_loss_weight,
+        use_cutout_augmentations=use_cutout_augmentations,
     )
 
     ### Load model -------------------------------------------------------------
@@ -189,6 +191,9 @@ def generate_image(
             "tv_loss_weight": tv_loss_weight,
         }
 
+        if use_cutout_augmentations:
+            details["use_cutout_augmentations"] = True
+
         if "git" in sys.modules:
             try:
                 repo = git.Repo(search_parent_directories=True)
@@ -257,6 +262,9 @@ def generate_image(
             "tv_loss_weight": tv_loss_weight,
         }
 
+        if use_cutout_augmentations:
+            details["use_cutout_augmentations"] = True
+
         if "git" in sys.modules:
             try:
                 repo = git.Repo(search_parent_directories=True)
@@ -453,6 +461,12 @@ def generate_image(
         else:
             tv_loss_weight = 0
 
+        use_cutout_augmentations = st.sidebar.checkbox(
+            "Use cutout augmentations",
+            value=True,
+            help="Increases image quality, uses additional 1-2 GiB of GPU memory",
+        )
+
         submitted = st.form_submit_button("Run!")
         # End of form
 
@@ -519,6 +533,7 @@ def generate_image(
             mse_weight=mse_weight,
             mse_weight_decay=mse_weight_decay,
             mse_weight_decay_steps=mse_weight_decay_steps,
+            use_cutout_augmentations=use_cutout_augmentations,
         )
         vid_display_slot.video("temp.mp4")
         # debug_slot.write(st.session_state) # DEBUG

logic.py

@@ -18,6 +18,7 @@
 from torch.nn import functional as F
 from torch import optim
 from torchvision import transforms
+import kornia.augmentation as K
 
 
 class Run:
@@ -70,7 +71,8 @@ def __init__(
         mse_weight=0.5,
         mse_weight_decay=0.1,
         mse_weight_decay_steps=50,
-        tv_loss_weight=1e-3
+        tv_loss_weight=1e-3,
+        use_cutout_augmentations: bool = True
         # use_augs: bool = True,
         # noise_fac: float = 0.1,
         # use_noise: Optional[float] = None,
@@ -130,6 +132,8 @@ def __init__(
         self.mse_weight_decay = mse_weight_decay
         self.mse_weight_decay_steps = mse_weight_decay_steps
 
+        self.use_cutout_augmentations = use_cutout_augmentations
+
         # For TV loss
         self.tv_loss_weight = tv_loss_weight
 
@@ -159,9 +163,28 @@ def model_init(self, init_image: Image.Image = None) -> None:
         cut_size = self.perceptor.visual.input_resolution
         e_dim = self.model.quantize.e_dim
         f = 2 ** (self.model.decoder.num_resolutions - 1)
+
+        if self.use_cutout_augmentations:
+            noise_fac = 0.1
+            augs = nn.Sequential(
+                K.RandomHorizontalFlip(p=0.5),
+                K.RandomSharpness(0.3, p=0.4),
+                K.RandomAffine(degrees=30, translate=0.1, p=0.8, padding_mode="border"),
+                K.RandomPerspective(0.2, p=0.4),
+                K.ColorJitter(hue=0.01, saturation=0.01, p=0.7),
+            )
+        else:
+            noise_fac = None
+            augs = None
+
         self.make_cutouts = MakeCutouts(
-            cut_size, self.args.cutn, cut_pow=self.args.cut_pow
+            cut_size,
+            self.args.cutn,
+            cut_pow=self.args.cut_pow,
+            noise_fac=noise_fac,
+            augs=augs,
         )
+
         n_toks = self.model.quantize.n_e
         toksX, toksY = self.args.size[0] // f, self.args.size[1] // f
         sideX, sideY = toksX * f, toksY * f

vqgan_utils.py

@@ -143,11 +143,13 @@ def parse_prompt(prompt):
 
 
 class MakeCutouts(nn.Module):
-    def __init__(self, cut_size, cutn, cut_pow=1.0):
+    def __init__(self, cut_size, cutn, cut_pow=1.0, noise_fac=None, augs=None):
         super().__init__()
         self.cut_size = cut_size
         self.cutn = cutn
         self.cut_pow = cut_pow
+        self.noise_fac = noise_fac
+        self.augs = augs
 
     def forward(self, input):
         sideY, sideX = input.shape[2:4]
@@ -162,7 +164,17 @@ def forward(self, input):
             offsety = torch.randint(0, sideY - size + 1, ())
             cutout = input[:, :, offsety : offsety + size, offsetx : offsetx + size]
             cutouts.append(resample(cutout, (self.cut_size, self.cut_size)))
-        return clamp_with_grad(torch.cat(cutouts, dim=0), 0, 1)
+
+        if self.augs:
+            batch = self.augs(torch.cat(cutouts, dim=0))
+        else:
+            batch = torch.cat(cutouts, dim=0)
+
+        if self.noise_fac:
+            facs = batch.new_empty([self.cutn, 1, 1, 1]).uniform_(0, self.noise_fac)
+            batch = batch + facs * torch.randn_like(batch)
+
+        return clamp_with_grad(batch, 0, 1)
 
 
 def load_vqgan_model(config_path, checkpoint_path):