I am working on adding the option to save OCR training images then users can send me the training images so I can retrain the OCR model to improve accuracy. Most likely I will release a new version in a month or two.
If you want a full color option try the EmpireTech IPC-T549M-ALED-S2. This is what I have at my front door. It is not 4k but has a Min. Illumination of 0.0005 Lux@F1.0 (Color, 30IRE) like the EmpireTech IPC-Color4K-T...
If the model work well I will train a YOLOv8 version. We currently do not have a YOLOv8 .NET module, I am using the YOLOv5 .NET module and are getting better performance then the YOLOv8 module.
I am training a model right now that will identify the make/model of vehicles. I am not sure how good it will be most of the model years are about 10+ years old.
Link to Dataset I am using capstone-g7-car-detection Object Detection Dataset by Sudharsan Rengarajan
CLASSES
AM General Hummer SUV...
The ALPR module uses the license-plate model first to find the license plates in the images sent. If the module does not find any license plates it returns nothing found. If it finds a license plate then it crops the plate and runs OCR on the cropped plate and send the OCR results.
Below are the versions and test script (ChatGPT wrote the script)
Pillow 9.4.0
opencv-python 4.8.1.78
import time
import cv2
from PIL import Image
def resize_pil(image, output_width):
width_percent = (output_width / float(image.width))...
See the below link, this might fix the issue
https://stackoverflow.com/questions/43184887/dll-load-failed-error-when-importing-cv2#:~:text=Add%20a%20comment-,28,Recently%20I%20have,-faced%20the%20similar
Below are images from IPC-T5442T-ZE and Color4K-T after replacing the the IPC-T5442T-ZE with a Color4K-T.
IPC-T5442T-ZE Exposure set to 1/60
Color4K-T Exposure set to 1/120