darius-muc/YOLOv7_Window_Detection_without_Measurements

YOLOv7 Window Detection (Without Measurements)

A YOLOv7 model fine-tuned for detecting window frames in technical drawings (specifically trained to detect windows without considering measurement text association).

Model Description

This model is a fine-tuned YOLOv7 object detector specialized for the window manufacturing industry. It automatically detects and locates window frames in order documents, proposals, and technical drawings.

Unlike our sibling model "with Measurements", this version focuses purely on the window geometry/frame itself.

Model Type: Object Detection (YOLOv7)
Industry: Window Manufacturing & Installation
Task: Window boundary detection in technical drawings
Classes: 1 (window)

Training Data

The model was trained on a custom dataset of technical drawings.

• Source: windows-frames2 dataset run
• Image Types: Scanned technical drawings, floor plans, and technical construction documents

Training Method

Base Model: YOLOv7 (pre-trained weights: yolov7_training.pt)
Training Approach: Transfer learning / Fine-tuning
Configuration:

• Epochs: 300
• Batch Size: 8
• Image Size: 640×640
• Device: CPU
• Optimizer: SGD with momentum
• Learning Rate: 0.01 (with cosine annealing)

Final Performance Metrics:

• Precision: 0.925
• Recall: 0.981
• mAP@0.5: 0.976
• mAP@0.5:0.95: 0.888

How to Use

Installation

pip install torch torchvision
git clone https://github.com/WongKinYiu/yolov7.git
cd yolov7

Inference Code

import torch
from models.experimental import attempt_load
from utils.general import non_max_suppression, scale_coords
from utils.datasets import letterbox
import cv2
import numpy as np

# Load model
device = torch.device('cpu')  # or 'cuda:0' for GPU
model = attempt_load('windows_without_measurements_best.pt', map_location=device)
model.eval()

# Prepare image
img_path = 'your_drawing.jpg'
img0 = cv2.imread(img_path)
img = letterbox(img0, 640, stride=32)[0]
img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x640x640
img = np.ascontiguousarray(img)
img = torch.from_numpy(img).to(device)
img = img.float() / 255.0
if img.ndimension() == 3:
    img = img.unsqueeze(0)

# Inference
with torch.no_grad():
    pred = model(img)[0]
    pred = non_max_suppression(pred, 0.25, 0.45)

# Process detections
for det in pred:
    if len(det):
        det[:, :4] = scale_coords(img.shape[2:], det[:, :4], img0.shape).round()
        for *xyxy, conf, cls in det:
            label = f'window {conf:.2f}'
            print(f"Detected: {label} at {xyxy}")
            cv2.rectangle(img0, (int(xyxy[0]), int(xyxy[1])), 
                         (int(xyxy[2]), int(xyxy[3])), (0, 255, 0), 2)

License

This model is released under the GPL-3.0 License (inherited from YOLOv7).

Contact

Developed by: Modernaus Ugdymo Centras (MUC)
Website: https://muc.lt