Update app.py

app.py

@@ -22,8 +22,10 @@ class App:
 
     @staticmethod
     def on_step1_complete(input_img: str, input_pose_vid: str):
-        return [gr.Image(label="Input Image", value=input_img, type="filepath", scale=5),
-                gr.Video(label="Input Aligned Pose Video", value=input_pose_vid, scale=5)]
+        return [
+            gr.Image(label="Input Image", value=input_img, type="filepath", scale=5),
+            gr.Video(label="Input Aligned Pose Video", value=input_pose_vid, scale=5)
+        ]
 
     def musepose_demo(self):
         with gr.Blocks() as demo:
@@ -37,9 +39,31 @@ class App:
                         with gr.Column(scale=3):
                             vid_dance_output = gr.Video(label="Aligned Pose Output", scale=5, interactive=False)
                             vid_dance_output_demo = gr.Video(label="Aligned Pose Output Demo", scale=5)
-                        # Rest of the column setup remains the same
                         with gr.Column(scale=3):
-                            # Column settings remain the same
+                            with gr.Column():
+                                nb_detect_resolution = gr.Number(label="Detect Resolution", value=512, precision=0)
+                                nb_image_resolution = gr.Number(label="Image Resolution", value=720, precision=0)
+                                nb_align_frame = gr.Number(label="Align Frame", value=0, precision=0)
+                                nb_max_frame = gr.Number(label="Max Frame", value=300, precision=0)
+                            with gr.Row():
+                                btn_align_pose = gr.Button("ALIGN POSE", variant="primary")
+
+                with gr.Column():
+                    examples = [
+                        [os.path.join("examples", "dance.mp4"), os.path.join("examples", "ref.png"),
+                         512, 720, 0, 300]
+                    ]
+                    ex_step1 = gr.Examples(examples=examples,
+                                           inputs=[vid_dance_input, img_pose_input, nb_detect_resolution,
+                                                   nb_image_resolution, nb_align_frame, nb_max_frame],
+                                           outputs=[vid_dance_output, vid_dance_output_demo],
+                                           fn=self.pose_alignment_infer.align_pose,
+                                           cache_examples="lazy")
+
+                btn_align_pose.click(fn=self.pose_alignment_infer.align_pose,
+                                     inputs=[vid_dance_input, img_pose_input, nb_detect_resolution, nb_image_resolution,
+                                             nb_align_frame, nb_max_frame],
+                                     outputs=[vid_dance_output, vid_dance_output_demo])
 
                 with gr.TabItem('2: MusePose Inference'):
                     with gr.Row():
@@ -49,8 +73,34 @@ class App:
                         with gr.Column(scale=3):
                             vid_output = gr.Video(label="MusePose Output", scale=5)
                             vid_output_demo = gr.Video(label="MusePose Output Demo", scale=5)
-                        # Rest of the settings remains the same
-            return demo
+
+                        with gr.Column(scale=3):
+                            with gr.Column():
+                                weight_dtype = gr.Dropdown(label="Compute Type", choices=["fp16", "fp32"], value="fp16")
+                                nb_width = gr.Number(label="Width", value=512, precision=0)
+                                nb_height = gr.Number(label="Height", value=512, precision=0)
+                                nb_video_frame_length = gr.Number(label="Video Frame Length", value=300, precision=0)
+                                nb_video_slice_frame_length = gr.Number(label="Video Slice Frame Number", value=48, precision=0)
+                                nb_video_slice_overlap_frame_number = gr.Number(label="Video Slice Overlap Frame Number", value=4, precision=0)
+                                nb_cfg = gr.Number(label="CFG (Classifier Free Guidance)", value=3.5, precision=0)
+                                nb_seed = gr.Number(label="Seed", value=99, precision=0)
+                                nb_steps = gr.Number(label="DDIM Sampling Steps", value=20, precision=0)
+                                nb_fps = gr.Number(label="FPS (Frames Per Second)", value=-1, precision=0, info="Set to '-1' to use same FPS with pose's")
+                                nb_skip = gr.Number(label="SKIP (Frame Sample Rate = SKIP+1)", value=1, precision=0)
+                            with gr.Row():
+                                btn_generate = gr.Button("GENERATE", variant="primary")
+
+                btn_generate.click(fn=self.musepose_infer.infer_musepose,
+                                   inputs=[img_musepose_input, vid_pose_input, weight_dtype, nb_width, nb_height,
+                                           nb_video_frame_length, nb_video_slice_frame_length,
+                                           nb_video_slice_overlap_frame_number, nb_cfg, nb_seed, nb_steps, nb_fps,
+                                           nb_skip],
+                                   outputs=[vid_output, vid_output_demo])
+                vid_dance_output.change(fn=self.on_step1_complete,
+                                        inputs=[img_pose_input, vid_dance_output],
+                                        outputs=[img_musepose_input, vid_pose_input])
+
+        return demo
 
     @staticmethod
     def header():
@@ -60,7 +110,7 @@ class App:
                 <a href="https://github.com/jhj0517/MusePose-WebUI" target="_blank">MusePose WebUI</a>
             </h1>
             <p style="font-size: 18px;">
-                <strong>Note</strong>: This space now allows video input up to <strong>10 seconds</strong> because ZeroGPU limits the function runtime to 2 minutes. <br>
+                <strong>Note</strong>: This space only allows video input up to <strong>10 seconds</strong> because ZeroGPU limits the function runtime to 2 minutes.<br>
                 If you want longer video inputs, you have to run it locally. Click the link above and follow the README to try it locally.<br><br>
                 When you have completed the <strong>1: Pose Alignment</strong> process, go to <strong>2: MusePose Inference</strong> and click the "GENERATE" button.
             </p>