Add openpose model

app_base.py

@@ -122,6 +122,16 @@ def create_demo(model: Model) -> gr.Blocks:
             0,
             True,
         ],
+        [
+            "assets/people.jpg",
+            "A couple, 4k photo, highly detailed",
+            "openpose",
+            "Photographic",
+            5.0,
+            1.0,
+            0,
+            True,
+        ],
     ]
 
     with gr.Blocks() as demo:

assets/README.md

@@ -5,3 +5,4 @@ These images were from the following URL:
 - https://huggingface.co/Adapter/t2iadapter/resolve/main/figs_SDXLV1.0/org_lin.jpg
 - https://huggingface.co/Adapter/t2iadapter/resolve/main/figs_SDXLV1.0/org_mid.jpg
 - https://huggingface.co/Adapter/t2iadapter/resolve/main/figs_SDXLV1.0/org_zeo.jpg
+- https://huggingface.co/Adapter/t2iadapter/resolve/main/people.jpg

model.py

@@ -2,12 +2,14 @@ import gc
 import os
 from abc import ABC, abstractmethod
 
+import numpy as np
 import PIL.Image
 import torch
 from controlnet_aux import (
     CannyDetector,
     LineartDetector,
     MidasDetector,
+    OpenposeDetector,
     PidiNetDetector,
     ZoeDetector,
 )
@@ -83,6 +85,7 @@ ADAPTER_REPO_IDS = {
     "lineart": "TencentARC/t2i-adapter-lineart-sdxl-1.0",
     "depth-midas": "TencentARC/t2i-adapter-depth-midas-sdxl-1.0",
     "depth-zoe": "TencentARC/t2i-adapter-depth-zoe-sdxl-1.0",
+    "openpose": "TencentARC/t2i-adapter-openpose-sdxl-1.0",
     # "recolor": "TencentARC/t2i-adapter-recolor-sdxl-1.0",
 }
 ADAPTER_NAMES = list(ADAPTER_REPO_IDS.keys())
@@ -114,7 +117,8 @@ class LineartPreprocessor(Preprocessor):
         self.model = LineartDetector.from_pretrained("lllyasviel/Annotators")
 
     def to(self, device: torch.device | str) -> Preprocessor:
-        return self.model.to(device)
+        self.model.to(device)
+        return self
 
     def __call__(self, image: PIL.Image.Image) -> PIL.Image.Image:
         return self.model(image, detect_resolution=384, image_resolution=1024)
@@ -127,18 +131,35 @@ class MidasPreprocessor(Preprocessor):
         )
 
     def to(self, device: torch.device | str) -> Preprocessor:
-        return self.model.to(device)
+        self.model.to(device)
+        return self
 
     def __call__(self, image: PIL.Image.Image) -> PIL.Image.Image:
         return self.model(image, detect_resolution=512, image_resolution=1024)
 
 
+class OpenposePreprocessor(Preprocessor):
+    def __init__(self):
+        self.model = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
+
+    def to(self, device: torch.device | str) -> Preprocessor:
+        self.model.to(device)
+        return self
+
+    def __call__(self, image: PIL.Image.Image) -> PIL.Image.Image:
+        out = self.model(image, detect_resolution=512, image_resolution=1024)
+        out = np.array(out)[:, :, ::-1]
+        out = PIL.Image.fromarray(np.uint8(out))
+        return out
+
+
 class PidiNetPreprocessor(Preprocessor):
     def __init__(self):
         self.model = PidiNetDetector.from_pretrained("lllyasviel/Annotators")
 
     def to(self, device: torch.device | str) -> Preprocessor:
-        return self.model.to(device)
+        self.model.to(device)
+        return self
 
     def __call__(self, image: PIL.Image.Image) -> PIL.Image.Image:
         return self.model(image, detect_resolution=512, image_resolution=1024, apply_filter=True)
@@ -159,7 +180,8 @@ class ZoePreprocessor(Preprocessor):
         )
 
     def to(self, device: torch.device | str) -> Preprocessor:
-        return self.model.to(device)
+        self.model.to(device)
+        return self
 
     def __call__(self, image: PIL.Image.Image) -> PIL.Image.Image:
         return self.model(image, gamma_corrected=True, image_resolution=1024)
@@ -175,6 +197,7 @@ if PRELOAD_PREPROCESSORS_IN_GPU_MEMORY:
         "lineart": LineartPreprocessor().to(device),
         "depth-midas": MidasPreprocessor().to(device),
         "depth-zoe": ZoePreprocessor().to(device),
+        "openpose": OpenposePreprocessor().to(device),
         "recolor": RecolorPreprocessor().to(device),
     }
 
@@ -188,6 +211,7 @@ elif PRELOAD_PREPROCESSORS_IN_CPU_MEMORY:
         "lineart": LineartPreprocessor(),
         "depth-midas": MidasPreprocessor(),
         "depth-zoe": ZoePreprocessor(),
+        "openpose": OpenposePreprocessor(),
         "recolor": RecolorPreprocessor(),
     }
 
@@ -207,6 +231,8 @@ else:
             return MidasPreprocessor()
         elif adapter_name == "depth-zoe":
             return ZoePreprocessor()
+        elif adapter_name == "openpose":
+            return OpenposePreprocessor()
         elif adapter_name == "recolor":
             return RecolorPreprocessor()
         else: