deep-anpr icon indicating copy to clipboard operation
deep-anpr copied to clipboard

can detect more than seven numbers?

Open willduan opened this issue 8 years ago • 4 comments

i'm very appreciate to know the project? I just want to use to detect some numbers and chracters? can it detect more than seven numbers? Thanks

willduan avatar Dec 01 '16 06:12 willduan

I am not sure about the performance. As a glance, that is possible. By adding more digits (common.py), the artificially made training images would have more digits. The network used in here should be extended to accept more digits. The easiest way is to increase the dimension of last FC output.

sapeyes avatar Jan 09 '17 05:01 sapeyes

Yes ofcourse it can, but the code modification doesn't involve changing common.py , but you have to change gen.py , the generate code function

sourabh2k15 avatar Apr 01 '17 13:04 sourabh2k15

@sourabh2k15 I've tried that, and I don't believe it works. I think you do have to modify code elsewhere other than gen.py because I believe especially the output of the fully connected layer thinks there are only 7 classes. It would be great if anyone figures an easy way to change the number of characters the nn expects.

josharnoldjosh avatar May 30 '17 23:05 josharnoldjosh

I had done it successfully, Indian license plates have 10 length. so i changed in gen.py the pattern , got test data , then changed model.py to change 7 to 10 everywhere, you also need to change it in train.py . Also finally when you get training weights you will want to run it on new images using detect.py, there also change 7s to 10s . Hope that helps.

Thanks.

On Wed, May 31, 2017 at 4:53 AM, Josh Arnold [email protected] wrote:

@sourabh2k15 https://github.com/sourabh2k15 I've tried that, and I don't believe it works. I think you do have to modify code elsewhere other than gen.py because I believe especially the output of the fully connected layer thinks there are only 7 classes. It would be great if anyone figures an easy way to change the number of characters the nn expects.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/matthewearl/deep-anpr/issues/16#issuecomment-305036956, or mute the thread https://github.com/notifications/unsubscribe-auth/AH09QlAR8A-VrizIID-UT-gGwQG8irFgks5r_KUEgaJpZM4LBCb0 .

sourabh2k15 avatar May 31 '17 06:05 sourabh2k15