Versión para permitir video 4K

Dockerfile

@@ -0,0 +1,34 @@
+# Use an official PyTorch image with CUDA support
+FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
+
+# Set working directory
+WORKDIR /app
+
+# Install system dependencies required for OpenCV and FFmpeg
+RUN apt-get update && apt-get install -y \
+    git \
+    ffmpeg \
+    libgl1-mesa-glx \
+    libglib2.0-0 \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements file
+COPY requirements.txt .
+
+# Install Python dependencies
+# Note: PySide6 and pyqtdarktheme are excluded/ignored if they fail, 
+# as they are for desktop GUI and not needed for this web demo.
+RUN pip install --no-cache-dir -r requirements.txt || true
+
+# Copy the rest of the application code
+COPY . .
+
+# Expose the port defined in app.py (default 8000)
+EXPOSE 8000
+
+# Set environment variables
+ENV GRADIO_SERVER_NAME="0.0.0.0"
+
+# Command to run the application
+# We use the --port argument to match the EXPOSE instruction
+CMD ["python", "hugging_face/app.py", "--port", "8000"]

hugging_face/app.py

@@ -132,7 +132,7 @@ def get_frames_from_video(video_input, video_state):
             if ret == True:
                 current_memory_usage = psutil.virtual_memory().percent
                 frames.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
-                if current_memory_usage > 90:
+                if current_memory_usage > 98:
                     break
             else:
                 break
@@ -141,14 +141,14 @@ def get_frames_from_video(video_input, video_state):
     image_size = (frames[0].shape[0],frames[0].shape[1]) 
 
     # resize if resolution too big
-    if image_size[0]>=1280 and image_size[0]>=1280:
-        scale = 1080 / min(image_size)
-        new_w = int(image_size[1] * scale)
-        new_h = int(image_size[0] * scale)
-        # update frames
-        frames = [cv2.resize(f, (new_w, new_h), interpolation=cv2.INTER_AREA) for f in frames]
-        # update image_size
-        image_size = (frames[0].shape[0],frames[0].shape[1]) 
+    # if image_size[0]>=1280 and image_size[0]>=1280:
+    #     scale = 1080 / min(image_size)
+    #     new_w = int(image_size[1] * scale)
+    #     new_h = int(image_size[0] * scale)
+    #     # update frames
+    #     frames = [cv2.resize(f, (new_w, new_h), interpolation=cv2.INTER_AREA) for f in frames]
+    #     # update image_size
+    #     image_size = (frames[0].shape[0],frames[0].shape[1]) 
 
     # initialize video_state
     video_state = {
@@ -363,8 +363,8 @@ def generate_video_from_frames(frames, output_path, fps=30, gray2rgb=False, audi
     video_temp_path = output_path.replace(".mp4", "_temp.mp4")
     
     # resize back to ensure input resolution
-    imageio.mimwrite(video_temp_path, frames, fps=fps, quality=7, 
-                     codec='libx264', ffmpeg_params=["-vf", f"scale={w}:{h}"])
+    imageio.mimwrite(video_temp_path, frames, fps=fps, 
+                     codec='libx264', ffmpeg_params=["-crf", "18", "-preset", "slow", "-vf", f"scale={w}:{h}"])
     
     # add audio to video if audio path exists
     if audio_path != "" and os.path.exists(audio_path):
@@ -425,6 +425,8 @@ from matanyone.model.matanyone import MatAnyone
 matanyone_model = MatAnyone.from_pretrained("PeiqingYang/MatAnyone")
 
 matanyone_model = matanyone_model.to(args.device).eval()
+# Force no internal resizing for high quality
+matanyone_model.cfg.max_internal_size = -1
 matanyone_processor = InferenceCore(matanyone_model, cfg=matanyone_model.cfg)
 
 # download test samples
@@ -450,7 +452,7 @@ description = r"""
 🔥 MatAnyone is a practical human video matting framework supporting target assignment 🎯.<br>
 🎪 Try to drop your video/image, assign the target masks with a few clicks, and get the the matting results 🤡!<br>
 
-*Note: Due to the online GPU memory constraints, any input with too big resolution will be resized to 1080p.<br>*
+*Note: High resolution inputs (4K) are supported but require significant RAM and VRAM.<br>*
 🚀 <b> If you encounter any issue (e.g., frozen video output) or wish to run on higher resolution inputs, please consider <u>duplicating this space</u> or 
 <u>launching the <a href='https://github.com/pq-yang/MatAnyone?tab=readme-ov-file#-interactive-demo' target='_blank'>demo</a> locally</u> following the GitHub instructions.</b>
 """
@@ -978,4 +980,4 @@ with gr.Blocks(theme=gr.themes.Monochrome(), css=my_custom_css) as demo:
     gr.Markdown(article)
 
 demo.queue()
-demo.launch(debug=True)
+demo.launch(debug=True, server_name="0.0.0.0", server_port=args.port)