tensorflow implementation of LPRnet. A lightweight deep network for number plate recognition.
- multiple scale CNN features
- CTC for variable length chars
- no RNN layers
generate plate images for training
python gen_plates.py
generate validation images
python gen_plates.py -s .\valid -n 200
train
python main.py -m train
or train with runtime-generated images
python main.py -m train -r
model checkpoint will be save for each SAVE_STEPS steps.
validation will be perform for each VALIDATE_EPOCHS epochs.
generate test images
python gen_plates.py -s .\test -n 200
restore checkpoint for test
python main.py -m test -c [checkpioint]
e.g
python main.py -m test -c .\checkpoint\LPRnet_steps8000_loss_0.069.ckpt
...
val loss: 0.31266
plate accuracy: 192-200 0.960, char accuracy: 1105-1115 0.99103
to test single image and show result
python main.py -m test -c [checkpoint] --img [image fullpath]
e.g
python main.py -m test -c .\checkpoint\LPRnet_steps5000_loss_0.215.ckpt --img .\test\AW73RHW_18771.jpg
...
restore from checkpoint: .\checkpoint\LPRnet_steps5000_loss_0.215.ckpt
AM73RHW
change TRAIN_DIR, VAL_DIR in LPRnet.py to folder contains your training/validation data.
image filename with the format [label]_XXXX
e.g AB12CD_0000.jpg
-
char set
change
CHARSif possible chars in label is different with default. -
char length
default input resolution (94x24) has 24 timesteps in CTC layer.
if your data have more than 8 chars in images, perhaps use wider resolution for good performance.
e.g input width 128 has 32 timesteps in CTC layer.