docker

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

fix

f

fix

fix

fix

fix

more

more

fix

Dockerfile

@@ -0,0 +1,16 @@
+FROM xuehy93/hififace:1.0
+
+
+RUN apt update && apt install -y wget
+
+
+WORKDIR /
+
+RUN wget https://public.ph.files.1drv.com/y4m_El1_AyFLmGuZaPWOqkytzM4qYtDc3BvNNL99JV1OLCEkmD4RTQjtHEXZ0SAWb7UPLV1IPB0KO2rFlyGJaV_kITLbuAHzJ73GwR_cgvXpkIGywaTnKsKVV1jJe1LoFcl7XsxatyGpaC8-Gupq6jjBnaqSBH4dgfYAmzUk8Wqiiuj_ml2duU7No0M1T426y3RqOJsqVHXEMVfV0B6HjzQFKCCZIgfHjjHvLIB3B3xP8Q?AVOverride=1 -O checkpoints.tar.gz
+
+RUN tar xfz checkpoints.tar.gz
+
+WORKDIR /app
+ADD ./ /app
+RUN chmod +x ./server.sh
+CMD ["./server.sh"]

README.md

@@ -1,8 +1,8 @@
 ---
 title: HiFiFace Inference
 emoji: 📉
-colorFrom: blue
-colorTo: yellow
+colorFrom: indigo
+colorTo: indigo
 sdk: docker
 pinned: false
 license: mit

app.py

@@ -9,7 +9,7 @@ from models.model import HifiFace
 
 
 class ConfigPath:
-    face_detector_weights = "./checkpoints/face_detector/face_detector_scrfd_10g_bnkps.onnx"
+    face_detector_weights = "/checkpoints/face_detector/face_detector_scrfd_10g_bnkps.onnx"
     model_path = ""
     model_idx = 80000
     ffmpeg_device = "cuda"
@@ -21,7 +21,7 @@ def main():
     parser = argparse.ArgumentParser(
         prog="benchmark", description="What the program does", epilog="Text at the bottom of help"
     )
-    parser.add_argument("-m", "--model_path", default="./checkpoints/standard_model")
+    parser.add_argument("-m", "--model_path", default="/checkpoints/hififace_pretrained/standard_model")
     parser.add_argument("-i", "--model_idx", default="320000")
     parser.add_argument("-f", "--ffmpeg_device", default="cpu")
     parser.add_argument("-d", "--device", default="cpu")
@@ -34,25 +34,29 @@ def main():
     cfg.device = args.device
     opt = TrainConfig()
     checkpoint = (cfg.model_path, cfg.model_idx)
+    model_path_1 = "/checkpoints/hififace_pretrained/with_gaze_and_mouth"
+    checkpoint1 = ("/checkpoints/hififace_pretrained/with_gaze_and_mouth", "190000")
     model = HifiFace(opt.identity_extractor_config, is_training=False, device=cfg.device, load_checkpoint=checkpoint)
 
+    model1 = HifiFace(opt.identity_extractor_config, is_training=False, device=cfg.device, load_checkpoint=checkpoint1)
     image_infer = ImageSwap(cfg, model)
-    video_infer = VideoSwap(cfg, model)
-
+    image_infer1 = ImageSwap(cfg, model1)
     def inference_image(source_face, target_face, shape_rate, id_rate, iterations):
         return image_infer.inference(source_face, target_face, shape_rate, id_rate, int(iterations))
 
-    def inference_video(source_face, target_video, shape_rate, id_rate, iterations):
-        return video_infer.inference(source_face, target_video, shape_rate, id_rate, int(iterations))
+    def inference_image1(source_face, target_face, shape_rate, id_rate, iterations):
+        return image_infer1.inference(source_face, target_face, shape_rate, id_rate, int(iterations))
 
     model_name = cfg.model_path.split("/")[-1] + ":" + f"{cfg.model_idx}"
+    model_name1 = model_path_1.split("/")[-1] + ":" + "190000"
     with gr.Blocks(title="FaceSwap") as demo:
         gr.Markdown(
             f"""
-        ### model: {model_name}
+        ### standard model: {model_name} \n
+        ### model with eye and mouth hm loss: {model_name1}
         """
         )
-        with gr.Tab("Image swap"):
+        with gr.Tab("Image swap with standard model"):
             with gr.Row():
                 source_image = gr.Image(shape=None, label="source image")
                 target_image = gr.Image(shape=None, label="target image")
@@ -88,10 +92,10 @@ def main():
                 outputs=output_image,
             )
 
-        with gr.Tab("Video swap"):
+        with gr.Tab("Image swap with eye&mouth hm loss model"):
             with gr.Row():
                 source_image = gr.Image(shape=None, label="source image")
-                target_video = gr.Video(value=None, label="target video")
+                target_image = gr.Image(shape=None, label="target image")
             with gr.Row():
                 with gr.Column():
                     structure_sim = gr.Slider(
@@ -115,22 +119,15 @@ def main():
                         step=1,
                         label="iters",
                     )
-                    video_btn = gr.Button("video swap")
-                output_video = gr.Video(value=None, label="Result")
-
-            video_btn.click(
-                fn=inference_video,
-                inputs=[
-                    source_image,
-                    target_video,
-                    structure_sim,
-                    id_sim,
-                    iters,
-                ],
-                outputs=output_video,
-            )
+                    image_btn = gr.Button("image swap")
+                output_image = gr.Image(shape=None, label="Result")
 
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+            image_btn.click(
+                fn=inference_image1,
+                inputs=[source_image, target_image, structure_sim, id_sim, iters],
+                outputs=output_image,
+            )
+        demo.launch(server_name="0.0.0.0", server_port=7860)
 
 
 if __name__ == "__main__":

configs/train_config.py

@@ -30,10 +30,10 @@ class TrainConfig:
     load_checkpoint = None  # ("/data/checkpoints/hififace/rebuilt_discriminator_SFF_c256_1683367464544", 400000)
 
     identity_extractor_config = {
-        "f_3d_checkpoint_path": "./checkpoints/Deep3DFaceRecon/epoch_20_new.pth",
-        "f_id_checkpoint_path": "./checkpoints/arcface/ms1mv3_arcface_r100_fp16_backbone.pth",
-        "bfm_folder": "./checkpoints/useful_ckpt/BFM",
-        "hrnet_path": "./checkpoints/useful_ckpt/face_98lmks/HR18-WFLW.pth",
+        "f_3d_checkpoint_path": "/checkpoints/Deep3DFaceRecon/epoch_20_new.pth",
+        "f_id_checkpoint_path": "/checkpoints/arcface/ms1mv3_arcface_r100_fp16_backbone.pth",
+        "bfm_folder": "/checkpoints/useful_ckpt/BFM",
+        "hrnet_path": "/checkpoints/useful_ckpt/face_98lmks/HR18-WFLW.pth",
     }
 
     visualize_interval: int = 100

data_process/model.py

@@ -1,4 +1,3 @@
-#!/usr/bin/python
 # -*- encoding: utf-8 -*-
 from typing import Tuple
 

data_process/resnet.py

@@ -1,4 +1,3 @@
-#!/usr/bin/python
 # -*- encoding: utf-8 -*-
 import torch
 import torch.nn as nn

server.sh

@@ -0,0 +1 @@
+python3 app.py