Successfully Trained YOLOv5 for Egg Counting: Need Advice on Real-Time Detection
Hey everyone! I wanted to give you an update on my egg-counting project. I decided to train a model using a dataset from Roboflow and chose YOLOv5 as the model. After setting up all the necessary dependencies in VS Code, I started training, but it was taking a really long time. So, I switched to Google Colab for its free GPU support, which turned out to be a great decision, the training only took 20 minutes!
I tested a few different egg images, and they were successfully detected and counted! My next goal is to implement real-time detection. Does anyone have any suggestions on the best approach for that?
13 Replies
I didn't realize Google Colab had free GPU access... what's the limitation on those resources?
I believe it is limited. They said something about their limitations fluctuating based on demand, not sure if thats still true.
interesting can we see some images you used for training?
Reliable.
Yes it have limitations specialy if you have a very big data to train, sessions can last up to 12 hours, and sometimes GPU availability depends on demand. If you use it a lot, you might hit some usage limits too. You can always upgrade to Colab Pro
Yes @ZacckOsiemo i download this data from roboflow website and it looks like this
It have 850 image , i used 577 for the training and 167 for validation and 106 for testing
nice but what about considering a specific scenario eggs on a tray
what is the accuracy?
Yes , that good idea I will try to add some to the dataset
achieved 91% precision and 89% [email protected] on the test set, i think it's fine
more than 90% is good
Yeah