Generate Synthetic Korea License Plates.
-
You can use this generator when there is insufficient data on the license plate.
-
I recommend pre-training with synthetic images and fine-tune with real data.
-
You can create synthetic license plate pictures by selecting the plate of the desired type.
-
The name of the photo shows the letters and numbers on the license plate.
-
Hangul in the plate was translated into English with the following rules.
-
Region : 서울 -> A, 인천 -> B ...
-
Hangul : 나 -> sk, 너 -> sj, 다 -> ek, 도 -> eh ...
Ex)
Type 1 : Z13wn0965
Type 2 : Z22aj0246
Type 3 : B16an7086
Type 4 : A48sk2287
Type 5 : Z19tn7921
Type 6 : 112ah0833
Type 7 : X50fk9747
File | Description |
---|---|
generator_original.py | generate images without any image distortion/augmentation. |
generator_augmentation.py | generate images with image augmentations such as random brightness. |
generator_perspective.py | generate images with perspective transform. |
generator_noise.py | generate images with noise such as rainfall, sunlight, shadows. |
-
Use generated license plates to create new car images.
-
You can use this generator when there is insufficient data.
-
I recommend pre-training with synthetic images and fine-tune with real data.
-
You need your own car images for this part.
File | Description |
---|---|
buildcar_click_one.py | click on the 4 corners of the license plate location to generate cars with new license plates. |
buildcar_click_multiple.py | same as above but for creating multiple images. |
buildcar_nonclick.py | no need to click (need the location info beforehand). |
File | Description |
---|---|
plate_detection.py | detect license plate and crop. |
- Align license plates using projective transform
File | Description |
---|---|
projective_transform.py | align license plates (need to know the 4 corner coordinates). |
- Reference: https://github.com/JaidedAI/EasyOCR
File | Description |
---|---|
recognize.py | use OCR to recognize Korean and numbers on license plates. |