app.py

import gradio as gr
import matplotlib.pyplot as plt
from PIL import Image
from ultralyticsplus import YOLO
import cv2
import numpy as np
from transformers import pipeline
import requests
from io import BytesIO
import os

model = YOLO('best (1).pt')
model2 = pipeline('image-classification','Kaludi/csgo-weapon-classification')
name = ['grenade','knife','pistol','rifle']
image_directory = "/home/user/app/image"
video_directory = "/home/user/app/video"

# url_example="https://drive.google.com/file/d/1bBq0bNmJ5X83tDWCzdzHSYCdg-aUL4xO/view?usp=drive_link"
# url_example='https://drive.google.com/uc?id=' + url_example.split('/')[-2]
# r = requests.get(url_example)
# im1 = Image.open(BytesIO(r.content))

# url_example="https://drive.google.com/file/d/16Z7QzvZ99fbEPj1sls_jOCJBsC0h_dYZ/view?usp=drive_link"
# url_example='https://drive.google.com/uc?id=' + url_example.split('/')[-2]
# r = requests.get(url_example)
# im2 = Image.open(BytesIO(r.content))

# url_example="https://drive.google.com/file/d/13mjTMS3eR0AKYSbV-Fpb3fTBno_T42JN/view?usp=drive_link"
# url_example='https://drive.google.com/uc?id=' + url_example.split('/')[-2]
# r = requests.get(url_example)
# im3 = Image.open(BytesIO(r.content))

# url_example="https://drive.google.com/file/d/1-XpFsa_nz506Ul6grKElVJDu_Jl3KZIF/view?usp=drive_link"
# url_example='https://drive.google.com/uc?id=' + url_example.split('/')[-2]
# r = requests.get(url_example)
# im4 = Image.open(BytesIO(r.content))
 # for i, r in enumerate(results):
      
 #    # Plot results image
 #      im_bgr = r.plot()  
 #      im_rgb = im_bgr[..., ::-1]  # Convert BGR to RGB


def response2(image: gr.Image = None,image_size: gr.Slider = 640, conf_threshold: gr.Slider = 0.3, iou_threshold: gr.Slider = 0.6):

    results = model.predict(image, conf=conf_threshold, iou=iou_threshold, imgsz=image_size
                           
    text = ""
    name_weap = ""
    
    box = results[0].boxes

    for r in results:
        im_array = r.plot()
        im = Image.fromarray(im_array[..., ::-1])

    
    
    for r in results:
        conf = np.array(r.boxes.conf)
        cls = np.array(r.boxes.cls)
        cls = cls.astype(int)
        xywh = np.array(r.boxes.xywh)
        xywh = xywh.astype(int)  
      
        for con, cl, xy in zip(conf, cls, xywh):
            cone = con.astype(float)
            conef = round(cone,3)
            conef = conef * 100
            text += (f"Detected {name[cl]} with confidence {round(conef,1)}% at ({xy[0]},{xy[1]})\n")
            
            if cl == 0:
                name_weap += name[cl] + '\n'
            elif cl == 1:
                name_weap += name[cl] + '\n'
            elif cl == 2:
                out = model2(image)
                name_weap += out[0]["label"] + '\n'
            elif cl == 3:
                out = model2(image)
                name_weap += out[0]["label"] + '\n'
      
    
    # xywh = int(results.boxes.xywh)
    # x = xywh[0]
    # y = xywh[1]
           
    return im, text, name_weap


inputs = [
    gr.Image(type="pil",  label="Input Image"),
    gr.Slider(minimum=320, maximum=1280, value=640,
                     step=32, label="Image Size"),
    gr.Slider(minimum=0.0, maximum=1.0, value=0.3,
                     step=0.05, label="Confidence Threshold"),
    gr.Slider(minimum=0.0, maximum=1.0, value=0.6,
                     step=0.05, label="IOU Threshold"),
]

outputs = [gr.Image( type="pil", label="Output Image"),
           gr.Textbox(label="Result"),
           gr.Textbox(label="Weapon Name")
          ]

examples = [[os.path.join(image_directory, "th (5).jpg"),640, 0.3, 0.6],
            [os.path.join(image_directory, "th (8).jpg"),640, 0.3, 0.6],
            [os.path.join(image_directory, "th (11).jpg"),640, 0.3, 0.6],
            [os.path.join(image_directory, "th (3).jpg"),640, 0.3, 0.6]
           ]
title = 'Weapon Detection Finetuned YOLOv8'
description = 'Image Size: Defines the image size for inference.\nConfidence Treshold: Sets the minimum confidence threshold for detections.\nIOU Treshold: Intersection Over Union (IoU) threshold for Non-Maximum Suppression (NMS). Useful for reducing duplicates.'


def pil_to_cv2(pil_image):
    open_cv_image = cv2.cvtColor(np.array(pil_image), cv2.COLOR_RGB2BGR)
    return open_cv_image


def process_video(video_path):
    cap = cv2.VideoCapture(video_path)
    
    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break
        
        pil_img = Image.fromarray(frame[..., ::-1])  
        result = model.predict(source=pil_img)
        for r in result:
            im_array = r.plot()
            processed_frame = Image.fromarray(im_array[..., ::-1])  
        yield processed_frame
    cap.release()


video_iface = gr.Interface(
    fn=process_video,
    inputs=[
        gr.Video(label="Upload Video", interactive=True)
    ],
    outputs=gr.Image(type="pil",label="Result"),
    title=title,
    description="Upload video for inference.",
    examples=[[os.path.join(video_directory, "ExampleRifle.mp4")],
        [os.path.join(video_directory, "Knife.mp4")],
    ]
)


image_iface = gr.Interface(fn=response2, inputs=inputs, outputs=outputs, examples=examples, title=title, description=description)

demo = gr.TabbedInterface([image_iface, video_iface], ["Image Inference", "Video Inference"])

if __name__ == '__main__':
    demo.launch()