Best ML technique for detecting multiple game cards in image
Best ML technique for detecting multiple game cards in image
Loading saved threads...
JanM · External communityPost link
External question — Cross Validated Stack Exchange
Author: JanM
Original post: https://stats.stackexchange.com/questions/303303
License: CC BY-SA 3.0 — https://creativecommons.org/licenses/by-sa/3.0/
Adaptation: HTML converted to plain text; contact email addresses removed.
This question is related to a new hobby project i want to start. I have some experience with ML techniques and neural networks, although only for regression problems as of now.
In my classification problem i get an image of multiple game cards, like those of a trading card game. I want to identify which card is shown automatically.
I have thought of
template matching
and
neural networks
.
Template matching
because the cards are exactly identical everytime, so i figured they are not hard to detect. But what changes is the angle and lighting of the pictures and i have heard template matching often fails for this.
Then i can use
neural networks
, which are very robust in those terms. I figure there are already some proven models existing, although something like AlexNet may be a bit too complex for my problem?
Is it advisable to use a neural network? If yes, what model should i use?
Or is there a better machine learning technique that suits my needs?
Quote
Report
J_Heads · External communityPost link
External answer — Cross Validated Stack Exchange
Author: J_Heads
Original post: https://stats.stackexchange.com/a/303363
License: CC BY-SA 3.0 — https://creativecommons.org/licenses/by-sa/3.0/
Adaptation: HTML converted to plain text; contact email addresses removed.
This is a very simple problem in terms of computer vision. I would recommend some processing to normalize the images and then trying something as simple nearest neighbors with mahalanobis distance.
A CNN will likely work well but comes with the need of massive data sets and computational power. As always in ML try the simple models before increasing your complexity.
Quote
Report
cdalitz · External communityPost link
External answer — Cross Validated Stack Exchange
Author: cdalitz
Original post: https://stats.stackexchange.com/a/675854
License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/
Adaptation: HTML converted to plain text; contact email addresses removed.
Using neural networks from scratch requires tons of training data until a reasonable recognition performance is achieved. A recent popular workaround is
transfer learning
: Apply a NN that has been pre-trained on millions of images and use its features (intermediate layers) for a simple kNN or SVN classifier. That way, you only need very little training data.
Popular pre-trained NNs for images are
DiNO
or
Franca
. They divide the images in patches and use a transformer architecture that not only yields feature vector for each patch, but also a global feature vector for the entire image. The distance measure is the cosine similarity (also known as
"attention"
):
Venkataramanan et al. "Franca: Nested matryoshka clustering for scalable visual representation learning." arXiv preprint arXiv:2507.14137 (2025)
Quote
Report
Post Reply
Quoted from Forex.com.bd-Editorial External question — Cross Validated Stack Exchange Author: JanM Source score (net votes, not local likes): 2 Original post: https://stats.stackexchange.com/questions/303303 License: CC BY-SA 3.0 — https://creativecommons.org/licenses/by-sa/3.0/ Adaptation: HTML converted to plain text; contact email addresses removed. This question is related to a new hobby project i want to start. I have some experience with ML techniques and neural networks, although only for regression problems as of now. In my classification problem i get an image of multiple game cards, like those of a trading card game. I want to identify which card is shown automatically. I have thought of template matching and neural networks . Template matching because the cards are exactly identical everytime, so i figured they are not hard to detect. But what changes is the angle and lighting of the pictures and i have heard template matching often fails for this. Then i can use neural networks , which are very robust in those terms. I figure there are already some proven models existing, although something like AlexNet may be a bit too complex for my problem? Is it advisable to use a neural network? If yes, what model should i use? Or is there a better machine learning technique that suits my needs?
Checking account access…