integrations/gradio/

[giscus[bot]]

integrations/gradio/

Learn to use Gradio and Ultralytics YOLOv8 for interactive object detection. Upload images and adjust detection parameters in real-time.

https://docs.ultralytics.com/integrations/gradio/

[SoulPerforms]

Hi, i'm new to this, I've finished training my yolov8 model and want to deploy it on Hugging Face Space with gradio. How to do video inference using gradio, so that i can input the video and make the model process it and display the output. Thanks

[Rifeet]

Hello @SoulPerforms,
if its okay for you can please share the interface code

[SoulPerforms]

you can go to files tab on my huggingface spaces and check the code

[Rifeet]

Thank you so much @SoulPerforms,
you saved my life

[lishihong-1]

Why does the video in the detection interface play slowly? What should I do to use the video stream detection in gradio?

[glenn-jocher]

Hey there! 👋 If your video playback in the Gradio interface is slower than expected, this might be due to the processing time required for each frame. Here are a couple of tips to help speed things up:

1. Optimize Model Inference: Reduce the image size for detection (imgsz), or you can try using a model optimized for speed like yolov8n.pt.

2. Adjust Gradio Code: If you're processing video streams, consider adding some buffering or optimizations in the frame handling to improve performance.

For real-time video stream detection, using Gradio can be a challenge since it doesn't naturally support streaming video feeds. I recommend employing a frame by frame approach where each frame is processed individually or integrating with another tool that supports video streaming.

# Simplified example assuming frame extraction is handled elsewhere
from PIL import Image
import gradio as gr
from ultralytics import YOLO

model = YOLO("yolov8n.pt")  # Load speed-optimized model

def process_frame(image):
    result = model.predict(image)
    processed_image = Image.fromarray(result[0].plot())
    return processed_image

iface = gr.Interface(fn=process_frame, inputs="image", outputs="image")
iface.launch()

This setup assumes you have a mechanism to feed video frames one at a time. Hope this helps! Let me know if you have any more questions. 😊