app.py

from ultralytics import YOLO
import supervision as sv
import cv2
import gradio as gr
import os
import numpy as np
from transformers import AutoProcessor, AutoModelForCausalLM
import torch
import requests
from PIL import Image
import glob
import pandas as pd
import time
import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)

device = 'cuda' if torch.cuda.is_available() else 'cpu'
model = AutoModelForCausalLM.from_pretrained("microsoft/Florence-2-base-ft", trust_remote_code=True).to(device).eval()
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-base-ft", trust_remote_code=True)
onnx_model = YOLO("models/best.onnx", task='detect')


def ends_with_number(s):
    return s[-1].isdigit()

def ocr(image, prompt="<OCR>"):
    original_height, original_width = image.shape[:2]
    inputs = processor(text=prompt, images=image, return_tensors="pt").to(device)
    generated_ids = model.generate(
        input_ids=inputs["input_ids"],
        pixel_values=inputs["pixel_values"],
        max_new_tokens=1024,
        early_stopping=False,
        do_sample=False,
        num_beams=3
    )
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]

    parsed_answer = processor.post_process_generation(
        generated_text,
        task=prompt,
        # image_size=(image.width, image.height)
        image_size=(original_width, original_height)
    )

    return parsed_answer

def parse_detection(detections):
    parsed_rows = []
    for i in range(len(detections.xyxy)):
        x_min = float(detections.xyxy[i][0])
        y_min = float(detections.xyxy[i][1])
        x_max = float(detections.xyxy[i][2])
        y_max = float(detections.xyxy[i][3])

        width = int(x_max - x_min)
        height = int(y_max - y_min)

        row = {
            "top": int(y_min),
            "left": int(x_min),
            "width": width,
            "height": height,
            "class_id": ""
            if detections.class_id is None
            else int(detections.class_id[i]),
            "confidence": ""
            if detections.confidence is None
            else float(detections.confidence[i]),
            "tracker_id": ""
            if detections.tracker_id is None
            else int(detections.tracker_id[i]),
        }

        if hasattr(detections, "data"):
            for key, value in detections.data.items():
                row[key] = (
                    str(value[i])
                    if hasattr(value, "__getitem__") and value.ndim != 0
                    else str(value)
                )
        parsed_rows.append(row)
    return parsed_rows


def cut_and_save_image(image, parsed_detections, output_dir):
    output_path_list = []

    for i, det in enumerate(parsed_detections):
        # Check if the class is 'mark'
        if det['class_name'] == 'mark':
            top = det['top']
            left = det['left']
            width = det['width']
            height = det['height']

            # Cut the image
            cut_image = image[top:top + height, left:left + width]
            # Save the image
            output_path = f"{output_dir}/cut_image_{i}.png"
            scaled_image = sv.scale_image(image=cut_image, scale_factor=4)
            cv2.imwrite(output_path, scaled_image, [int(cv2.IMWRITE_JPEG_QUALITY), 500])
            output_path_list.append(output_path)
    return output_path_list

def analysis(progress=gr.Progress()):
    progress(0, desc="Analyzing...")
    list_files = glob.glob("output/*.png")
    prompt = "<OCR>"
    results = {}
    for filepath in progress.tqdm(list_files):
        basename = os.path.basename(filepath)

        image = cv2.imread(filepath)
        
        start_time = time.time()
        parsed_answer = ocr(image, prompt)
        
        if not ends_with_number(parsed_answer[prompt]):
            parsed_answer[prompt] += "1"
        results[parsed_answer[prompt]] = results.get(parsed_answer[prompt], 0) + 1
        print(basename, parsed_answer[prompt])
        print("Time taken:", time.time() - start_time)
    return pd.DataFrame(results.items(), columns=['Mark', 'Total']).reset_index(drop=False).rename(columns={'index': 'No.'})

def inference(
    image_path,
    conf_threshold,
    iou_threshold,
):
    """
    YOLOv8 inference function
    Args:
        image_path: Path to the image
        conf_threshold: Confidence threshold
        iou_threshold: IoU threshold
    Returns:
        Rendered image
    """
    image = cv2.imread(image_path)
    original_height, original_width = image.shape[:2]
    print(image.shape)

    results = onnx_model(image, conf=conf_threshold, iou=iou_threshold)[0]
    detections = sv.Detections.from_ultralytics(results)
    parsed_detections = parse_detection(detections)
    output_dir = "output"
    # Check if the output directory exists, clear all the files inside
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)
    else:
        for f in os.listdir(output_dir):
            os.remove(os.path.join(output_dir, f))

    output_path_list = cut_and_save_image(image, parsed_detections, output_dir)

    box_annotator = sv.BoxAnnotator()
    label_annotator = sv.LabelAnnotator(text_position=sv.Position.TOP_LEFT, text_thickness=1, text_padding=2)
    annotated_image = image.copy()
    annotated_image = box_annotator.annotate(
        scene=annotated_image,
        detections=detections
    )
    annotated_image = label_annotator.annotate(scene=annotated_image, detections=detections)
    return annotated_image, output_path_list


TITLE = "<h1 style='font-size: 2.5em; text-align: center;'>Identify objects in construction design</h1>"
DESCRIPTION = """<p style='font-size: 1.5em; line-height: 1.6em; text-align: left;'>Welcome to the object 
identification application. This tool allows you to upload an image, and it will identify and annotate objects within 
the image. Additionally, you can perform OCR analysis on the detected objects.</p> """
CSS = """
  #output {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
  h1 {
    text-align: center;
  }
"""
EXAMPLES = [
    ['examples/train1.png', 0.6, 0.25],
    ['examples/train2.png', 0.9, 0.25],
    ['examples/train3.png', 0.6, 0.25]
]


with gr.Blocks(theme=gr.themes.Soft(), css=CSS) as demo:
    gr.HTML(TITLE)
    gr.HTML(DESCRIPTION)
    with gr.Tab(label="Identify objects"):
        with gr.Row():
            input_img = gr.Image(type="filepath", label="Upload Image")
            output_img = gr.Image(type="filepath", label="Output Image")
        with gr.Row():
            with gr.Column():
                conf_thres = gr.Slider(minimum=0.0, maximum=1.0, value=0.6, step=0.05, label="Confidence Threshold")
            with gr.Column():
                iou = gr.Slider(minimum=0.0, maximum=1.0, value=0.25, step=0.05, label="IOU Threshold")
        with gr.Row():
            with gr.Column():
                submit_btn = gr.Button(value="Predict")
            with gr.Column():
                analysis_btn = gr.Button(value="Analysis")
        with gr.Row():
            output_df = gr.Dataframe(label="Results")
        with gr.Row():
            with gr.Accordion("Gallery", open=False):
                gallery = gr.Gallery(label="Detected Mark Object", columns=3)
        submit_btn.click(inference, [input_img, conf_thres, iou], [output_img, gallery])
        analysis_btn.click(analysis, [], [output_df])
        examples = gr.Examples(
                    EXAMPLES,
                    fn=inference,
                    inputs=[input_img, conf_thres, iou],
                    outputs=[output_img, gallery],
                    cache_examples=False,
                )

demo.launch(debug=True)