-
Notifications
You must be signed in to change notification settings - Fork 7
Details and Errors
Harshit Dave edited this page Dec 22, 2021
·
2 revisions
- I used the object detection model because it's way more accurate and way more resilient than using traditional computer vision techniques.
- Verification text ran for 21 tests with approximately 17.079 secs.
- Here label map represents all the possible objects that the model will detect which is one object that is the license plate.
- Now, before starting with OCR, one should pip install Pytorch with CUDA acceleration, now it all depends on a particular's machine so a user can install **Pytorch with GPU (CUDA 11.3) **
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
or without GPUconda install pytorch torchvision torchaudio cpuonly -c pytorch
, this all is for windows machine. - It generally be wise to split the GPU because generally TensorFlow training model uses the whole GPU and then it slows our detection process.
- I used EasyOCR because it is better on numbers and has an accuracy of around 95%.
- For detecting license plates, generally the threshold is set to 0.8 for high accuracy.
- By default OpenCV works with the colour format [BGR] so one should use
cv2.COLOR_BGR2RGB
. - To save images with a unique name in the folder and in the .csv file, I used UUID that allows creating a unique uniform identifier.
- Average Precision (AP) is obtained by finding the area under the precision-recall curve. The mAP for object detection is the average of the AP calculated for all the classes to determine the accuracy of a set of object detections from a model when compared to ground-truth object annotations of a dataset.
- SSD MobileNet V2 FPNLite 320x320: speed - 22ms and COCO mAP - 22.2, got these information from tensorflow zoo model
- Faced an error while importing wget library.
- Faced an error while installing pycocotools. Error:
Value Error: numpy.ndarray size changed may indicate binary compatibility. Expected 88 from encoder, got from pyobject.
- While installing EasyOCR, my machine didn't support PyTorch with CUDA 11.3, so due to that I faced an error.