Depth Anything V2 Support (#2977)

* Add depth_anything_v2.py

* Update preprocessor_compiled.py

Added depth_anything_v2 preprocessor

* Update processor.py

Added depth_anything_v2 preprocessor

* Update depth_test.py

Added depth_anything_v2 to general_depth_modules

* Update modules_test.py

Added depth_anything_v2 to expected_module_names & expected_module_alias

* Update install.py

Added depth_anything_v2 to install.py

* Update README.md

* Fix model URL

---------

Co-authored-by: Mackinations.Ai <133395980+MackinationsAi@users.noreply.github.com>

README.md

@@ -7,6 +7,7 @@ This extension is for AUTOMATIC1111's [Stable Diffusion web UI](https://github.c
 
 # News
 
+- [2024-07-01] 🔥[v1.1.452] Depth Anything V2 - UDAV2 depth Preprocessor [Pull thread: https://github.com/Mikubill/sd-webui-controlnet/pull/2969]
 - [2024-05-19] 🔥[v1.1.449] Anyline Preprocessor & MistoLine SDXL model [Discussion thread: https://github.com/Mikubill/sd-webui-controlnet/discussions/2907]
 - [2024-05-04] 🔥[v1.1.447] PuLID [Discussion thread: https://github.com/Mikubill/sd-webui-controlnet/discussions/2841]
 - [2024-04-30] 🔥[v1.1.446] Effective region mask supported for ControlNet/IPAdapter [Discussion thread: https://github.com/Mikubill/sd-webui-controlnet/discussions/2831]

annotator/depth_anything_v2.py

@@ -0,0 +1,78 @@
+import os
+import torch
+import cv2
+import numpy as np
+import torch.nn.functional as F
+from torchvision.transforms import Compose
+from safetensors.torch import load_file
+
+from depth_anything_v2.dpt import DepthAnythingV2
+from depth_anything_v2.util.transform import Resize, NormalizeImage, PrepareForNet
+from .util import load_model
+from .annotator_path import models_path
+
+transform = Compose(
+    [
+        Resize(
+            width=518,
+            height=518,
+            resize_target=False,
+            keep_aspect_ratio=True,
+            ensure_multiple_of=14,
+            resize_method="lower_bound",
+            image_interpolation_method=cv2.INTER_CUBIC,
+        ),
+        NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+        PrepareForNet(),
+    ]
+)
+
+class DepthAnythingV2Detector:
+    """https://github.com/MackinationsAi/Upgraded-Depth-Anything-V2"""
+
+    model_dir = os.path.join(models_path, "depth_anything_v2")
+
+    def __init__(self, device: torch.device):
+        self.device = device
+        self.model = (
+            DepthAnythingV2(
+                encoder="vitl",
+                features=256,
+                out_channels=[256, 512, 1024, 1024],
+            )
+            .to(device)
+            .eval()
+        )
+        remote_url = os.environ.get(
+            "CONTROLNET_DEPTH_ANYTHING_V2_MODEL_URL",
+            "https://huggingface.co/MackinationsAi/Depth-Anything-V2_Safetensors/resolve/main/depth_anything_v2_vitl.safetensors",
+        )
+        model_path = load_model(
+            "depth_anything_v2_vitl.safetensors", remote_url=remote_url, model_dir=self.model_dir
+        )
+        self.model.load_state_dict(load_file(model_path))
+
+    def __call__(self, image: np.ndarray, colored: bool = True) -> np.ndarray:
+        self.model.to(self.device)
+        h, w = image.shape[:2]
+
+        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) / 255.0
+        image = transform({"image": image})["image"]
+        image = torch.from_numpy(image).unsqueeze(0).to(self.device)
+        @torch.no_grad()
+        def predict_depth(model, image):
+            return model(image)
+        depth = predict_depth(self.model, image)
+        depth = F.interpolate(
+            depth[None], (h, w), mode="bilinear", align_corners=False
+        )[0, 0]
+        depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
+        depth = depth.cpu().numpy().astype(np.uint8)
+        if colored:
+            depth_color = cv2.applyColorMap(depth, cv2.COLORMAP_INFERNO)[:, :, ::-1]
+            return depth_color
+        else:
+            return depth
+
+    def unload_model(self):
+        self.model.to("cpu")

install.py

@@ -167,6 +167,7 @@ try_install_from_wheel(
     ),
     version="2024.2.12.0",
 )
+
 try_install_from_wheel(
     "depth_anything",
     wheel_url=os.environ.get(
@@ -175,6 +176,14 @@ try_install_from_wheel(
     ),
 )
 
+try_install_from_wheel(
+    "depth_anything_v2",
+    wheel_url=os.environ.get(
+        "DEPTH_ANYTHING_V2_WHEEL",
+        "https://github.com/MackinationsAi/UDAV2-ControlNet/releases/download/v1.0.0/depth_anything_v2-2024.7.1.0-py2.py3-none-any.whl",
+    ),
+)
+
 try_install_from_wheel(
     "dsine",
     wheel_url=os.environ.get(

scripts/preprocessor/legacy/preprocessor_compiled.py

@@ -168,6 +168,22 @@ legacy_preprocessors = {
             "Depth"
         ]
     },
+    "depth_anything_v2": {
+        "label": "depth_anything_v2",
+        "call_function": functools.partial(depth_anything_v2, colored=False),
+        "unload_function": unload_depth_anything_v2,
+        "managed_model": "model_depth_anything_v2",
+        "model_free": False,
+        "no_control_mode": False,
+        "resolution": None,
+        "slider_1": None,
+        "slider_2": None,
+        "slider_3": None,
+        "priority": 0,
+        "tags": [
+            "Depth"
+        ]
+    },
     "depth_hand_refiner": {
         "label": "depth_hand_refiner",
         "call_function": g_hand_refiner_model.run_model,

scripts/preprocessor/legacy/processor.py

@@ -189,6 +189,25 @@ def unload_depth_anything():
         model_depth_anything.unload_model()
 
 
+model_depth_anything_v2 = None
+
+
+def depth_anything_v2(img, res:int = 512, colored:bool = True, **kwargs):
+    img, remove_pad = resize_image_with_pad(img, res)
+    global model_depth_anything_v2
+    if model_depth_anything_v2 is None:
+        with Extra(torch_handler):
+            from annotator.depth_anything_v2 import DepthAnythingV2Detector
+            device = devices.get_device_for("controlnet")
+            model_depth_anything_v2 = DepthAnythingV2Detector(device)
+    return remove_pad(model_depth_anything_v2(img, colored=colored)), True
+
+
+def unload_depth_anything_v2():
+    if model_depth_anything_v2 is not None:
+        model_depth_anything_v2.unload_model()
+
+
 model_midas = None
 
 

tests/web_api/full_coverage/depth_test.py

@@ -18,6 +18,7 @@ general_depth_modules = [
     "depth_leres",
     "depth_leres++",
     "depth_anything",
+    "depth_anything_v2",
 ]
 hand_refiner_module = "depth_hand_refiner"
 

tests/web_api/modules_test.py

@@ -15,6 +15,7 @@ expected_module_names = {
     "densepose_parula",
     "depth",
     "depth_anything",
+    "depth_anything_v2",
     "depth_hand_refiner",
     "depth_leres",
     "depth_leres++",
@@ -82,6 +83,7 @@ expected_module_alias = {
     "densepose (pruple bg & purple torso)",
     "densepose_parula (black bg & blue torso)",
     "depth_anything",
+    "depth_anything_v2",
     "depth_hand_refiner",
     "depth_leres",
     "depth_leres++",