Add normal stencil

apps/stable_diffusion/src/utils/stable_args.py

@@ -422,7 +422,7 @@ def is_valid_file(arg):
 
 p.add_argument(
     "--use_stencil",
-    choices=["canny", "openpose", "scribble"],
+    choices=["canny", "openpose", "scribble", "normal"],
     help="Enable the stencil feature.",
 )
 

apps/stable_diffusion/src/utils/stencils/__init__.py

@@ -1,2 +1,5 @@
 from apps.stable_diffusion.src.utils.stencils.canny import CannyDetector
 from apps.stable_diffusion.src.utils.stencils.openpose import OpenposeDetector
+from apps.stable_diffusion.src.utils.stencils.normal_bae import (
+    NormalBaeDetector,
+)

apps/stable_diffusion/src/utils/stencils/normal_bae/__init__.py

@@ -0,0 +1,140 @@
+import os
+import types
+import warnings
+
+import cv2
+import numpy as np
+import torch
+import torchvision.transforms as transforms
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from .nets.NNET import NNET
+
+
+def HWC3(x):
+    assert x.dtype == np.uint8
+    if x.ndim == 2:
+        x = x[:, :, None]
+    assert x.ndim == 3
+    H, W, C = x.shape
+    assert C == 1 or C == 3 or C == 4
+    if C == 3:
+        return x
+    if C == 1:
+        return np.concatenate([x, x, x], axis=2)
+    if C == 4:
+        color = x[:, :, 0:3].astype(np.float32)
+        alpha = x[:, :, 3:4].astype(np.float32) / 255.0
+        y = color * alpha + 255.0 * (1.0 - alpha)
+        y = y.clip(0, 255).astype(np.uint8)
+        return y
+
+
+# load model
+def load_checkpoint(fpath, model):
+    ckpt = torch.load(fpath, map_location="cpu")["model"]
+
+    load_dict = {}
+    for k, v in ckpt.items():
+        if k.startswith("module."):
+            k_ = k.replace("module.", "")
+            load_dict[k_] = v
+        else:
+            load_dict[k] = v
+
+    model.load_state_dict(load_dict)
+    return model
+
+
+class NormalBaeDetector:
+    def __init__(self, model):
+        self.model = model
+        self.norm = transforms.Normalize(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+        )
+
+    @classmethod
+    def from_pretrained(
+        cls, pretrained_model_or_path, filename=None, cache_dir=None
+    ):
+        filename = filename or "scannet.pt"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(
+                pretrained_model_or_path, filename, cache_dir=cache_dir
+            )
+
+        args = types.SimpleNamespace()
+        args.mode = "client"
+        args.architecture = "BN"
+        args.pretrained = "scannet"
+        args.sampling_ratio = 0.4
+        args.importance_ratio = 0.7
+        model = NNET(args)
+        model = load_checkpoint(model_path, model)
+        model.eval()
+
+        return cls(model)
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+
+    # def __call__(self, input_image, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+    def __call__(self, input_image, output_type="pil", **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn(
+                "return_pil is deprecated. Use output_type instead.",
+                DeprecationWarning,
+            )
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn(
+                "Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions"
+            )
+            if output_type:
+                output_type = "pil"
+
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        # input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        image_normal = input_image
+        with torch.no_grad():
+            image_normal = torch.from_numpy(image_normal).float().to(device)
+            image_normal = image_normal / 255.0
+            image_normal = rearrange(image_normal, "h w c -> 1 c h w")
+            image_normal = self.norm(image_normal)
+
+            normal = self.model(image_normal)
+            normal = normal[0][-1][:, :3]
+            # d = torch.sum(normal ** 2.0, dim=1, keepdim=True) ** 0.5
+            # d = torch.maximum(d, torch.ones_like(d) * 1e-5)
+            # normal /= d
+            normal = ((normal + 1) * 0.5).clip(0, 1)
+
+            normal = rearrange(normal[0], "c h w -> h w c").cpu().numpy()
+            normal_image = (normal * 255.0).clip(0, 255).astype(np.uint8)
+
+        detected_map = normal_image
+        detected_map = HWC3(detected_map)
+
+        # img = resize_image(input_image, image_resolution)
+        H, W, C = input_image.shape
+
+        detected_map = cv2.resize(
+            detected_map, (W, H), interpolation=cv2.INTER_LINEAR
+        )
+
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

apps/stable_diffusion/src/utils/stencils/normal_bae/nets/NNET.py

@@ -0,0 +1,22 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .submodules.encoder import Encoder
+from .submodules.decoder import Decoder
+
+
+class NNET(nn.Module):
+    def __init__(self, args):
+        super(NNET, self).__init__()
+        self.encoder = Encoder()
+        self.decoder = Decoder(args)
+
+    def get_1x_lr_params(self):  # lr/10 learning rate
+        return self.encoder.parameters()
+
+    def get_10x_lr_params(self):  # lr learning rate
+        return self.decoder.parameters()
+
+    def forward(self, img, **kwargs):
+        return self.decoder(self.encoder(img), **kwargs)

apps/stable_diffusion/src/utils/stencils/normal_bae/nets/baseline.py

@@ -0,0 +1,95 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .submodules.submodules import UpSampleBN, norm_normalize
+
+
+# This is the baseline encoder-decoder we used in the ablation study
+class NNET(nn.Module):
+    def __init__(self, args=None):
+        super(NNET, self).__init__()
+        self.encoder = Encoder()
+        self.decoder = Decoder(num_classes=4)
+
+    def forward(self, x, **kwargs):
+        out = self.decoder(self.encoder(x), **kwargs)
+
+        # Bilinearly upsample the output to match the input resolution
+        up_out = F.interpolate(
+            out,
+            size=[x.size(2), x.size(3)],
+            mode="bilinear",
+            align_corners=False,
+        )
+
+        # L2-normalize the first three channels / ensure positive value for concentration parameters (kappa)
+        up_out = norm_normalize(up_out)
+        return up_out
+
+    def get_1x_lr_params(self):  # lr/10 learning rate
+        return self.encoder.parameters()
+
+    def get_10x_lr_params(self):  # lr learning rate
+        modules = [self.decoder]
+        for m in modules:
+            yield from m.parameters()
+
+
+# Encoder
+class Encoder(nn.Module):
+    def __init__(self):
+        super(Encoder, self).__init__()
+
+        basemodel_name = "tf_efficientnet_b5_ap"
+        basemodel = torch.hub.load(
+            "rwightman/gen-efficientnet-pytorch",
+            basemodel_name,
+            pretrained=True,
+        )
+
+        # Remove last layer
+        basemodel.global_pool = nn.Identity()
+        basemodel.classifier = nn.Identity()
+
+        self.original_model = basemodel
+
+    def forward(self, x):
+        features = [x]
+        for k, v in self.original_model._modules.items():
+            if k == "blocks":
+                for ki, vi in v._modules.items():
+                    features.append(vi(features[-1]))
+            else:
+                features.append(v(features[-1]))
+        return features
+
+
+# Decoder (no pixel-wise MLP, no uncertainty-guided sampling)
+class Decoder(nn.Module):
+    def __init__(self, num_classes=4):
+        super(Decoder, self).__init__()
+        self.conv2 = nn.Conv2d(2048, 2048, kernel_size=1, stride=1, padding=0)
+        self.up1 = UpSampleBN(skip_input=2048 + 176, output_features=1024)
+        self.up2 = UpSampleBN(skip_input=1024 + 64, output_features=512)
+        self.up3 = UpSampleBN(skip_input=512 + 40, output_features=256)
+        self.up4 = UpSampleBN(skip_input=256 + 24, output_features=128)
+        self.conv3 = nn.Conv2d(
+            128, num_classes, kernel_size=3, stride=1, padding=1
+        )
+
+    def forward(self, features):
+        x_block0, x_block1, x_block2, x_block3, x_block4 = (
+            features[4],
+            features[5],
+            features[6],
+            features[8],
+            features[11],
+        )
+        x_d0 = self.conv2(x_block4)
+        x_d1 = self.up1(x_d0, x_block3)
+        x_d2 = self.up2(x_d1, x_block2)
+        x_d3 = self.up3(x_d2, x_block1)
+        x_d4 = self.up4(x_d3, x_block0)
+        out = self.conv3(x_d4)
+        return out