Videos

[waddahAldrobi]

Can you provide the video you used?

[geaxgx]

Do you mean the videos of the cards (one video par card) used to generate the data ?
It would only be helpful if you want to recognize cards from the same brand as the one I have used (unfortunately, it is not a famous brand like Bicycle). Otherwise, you won't get the same accuracy as demonstrated in the video. I suggest you to shoot new videos with the cards you used.

If you just want to test the scripts of the notebook, you can use the video file 'test/2c.avi'

[waddahAldrobi]

No I'm well aware of that, but I was trying to collect as many card types as possible, that's why I asked for them. If you can email me a zip, that would be great, if you don't want to add them to the repo. Also, does it matter where the card is located when I shoot my own video? I tried it and I felt like the location and size of the card were hard coded. Thanks, Waddah

…

On Sun, 10 Mar 2019 at 13:05, geaxgx ***@***.***> wrote: Do you mean the videos of the cards (one video par card) used to generate the data ? It would only be helpful if you want to recognize cards from the same brand as the one I have used (unfortunately, it is not a famous brand like Bicycle). Otherwise, you won't get the same accuracy as demonstrated in the video. I suggest you to shoot new videos with the cards you used. If you just want to test the scripts of the notebook, you can use the video file 'test/2c.avi' — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AbVaU3vl2aQ8Lv7qvUmgs1EYMoB98M7nks5vVTtrgaJpZM4banR0> .

[geaxgx]

Sorry, I currently don't have access to the videos you are asking for (they are on a separate drive in another location). But if you try to collect as many card types as possible, I would suggest to adapt the scripts in the notebook to deal with a picture of a card instead of a video of that card. It is much simpler. It just mean that you only take one picture of each card and rely on the data augmentation library to generate variations of that card. When I started this project, I wasn't aware that I could use the capabilities of the data augmentation library to simulate different lighting conditions, instead of a real external light.
What is hard coded is the dimensions of the card, not the location (see § C.1 in the notebook). So you have to modify these values when you work on a new card type. If you decide to take a picture of the cards, as I suggest, you want the best lighting condition, and sharp and neat pictures. The data augmentation library will artficially add noise and blurring.

[waddahAldrobi]

Ahh okay, sorry one last question.
Are the dimensions of the cards based on the actual size (i.e. using a ruler)
Or did you label them?

[geaxgx]

Yes, I used a ruler to get the actual size.