Change face_app detection size

lib/pipline_ConsistentID.py

@@ -12,7 +12,6 @@ from huggingface_hub.utils import validate_hf_hub_args
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipeline
-from diffusers.utils.import_utils import is_xformers_available
 from .functions import insert_markers_to_prompt, masks_for_unique_values, apply_mask_to_raw_image, tokenize_and_mask_noun_phrases_ends, prepare_image_token_idx
 from .functions import ProjPlusModel, masks_for_unique_values
 from .attention import Consistent_IPAttProcessor, Consistent_AttProcessor, FacialEncoder
@@ -39,11 +38,11 @@ class ConsistentIDPipeline(StableDiffusionPipeline):
         torch_device: Optional[Union[str, torch.device]] = None,
         torch_dtype: Optional[torch.dtype] = None,
     ):
-        super().to(torch_device,            torch_dtype)
-        self.bise_net.to(torch_device,      dtype=torch_dtype)
-        self.clip_encoder.to(torch_device,  dtype=torch_dtype)
-        self.image_proj_model.to(torch_device, dtype=torch_dtype)
-        self.FacialEncoder.to(torch_device, dtype=torch_dtype)
+        super().to(torch_device,                dtype=torch_dtype)
+        self.bise_net.to(torch_device,          dtype=torch_dtype)
+        self.clip_encoder.to(torch_device,      dtype=torch_dtype)
+        self.image_proj_model.to(torch_device,  dtype=torch_dtype)
+        self.FacialEncoder.to(torch_device,     dtype=torch_dtype)
         # If the unet is not released, the ip_layers should be moved to the specified device and dtype.
         if not isinstance(self.unet, edict):
             self.ip_layers.to(torch_device,     dtype=torch_dtype)
@@ -75,7 +74,8 @@ class ConsistentIDPipeline(StableDiffusionPipeline):
 
         # face_app: FaceAnalysis object
         self.face_app = FaceAnalysis(name="buffalo_l", root='models/insightface', providers=['CPUExecutionProvider'])
-        self.face_app.prepare(ctx_id=0, det_size=(640, 640))
+        # The original det_size=(640, 640) is too large and face_app often fails to detect faces.
+        self.face_app.prepare(ctx_id=0, det_size=(512, 512))
 
         if not os.path.exists(consistentID_weight_path):
             ### Download pretrained models