Question/Issue: Increase maximum training time for a project
**Project ID:**274744
Context/Use case:
I am a 14 year old student from Germany and for a student competition I have created a smart fridge which can automatically detect food. I won the first round of the competition and got the feedback that I should increase the number of different food types that my fridge can detect for the next round. I would really like to do this but I am already hitting the limit of max. 20 min training time with my current dataset. How can I temporarly increase this limit to > 60 min, so I can build a modell that supports more types of food? I would need to have the increased training time until mid of April, till then I plan to be finished with my project. Thanks for any help
I had the same issue a few weeks ago and the solution for me was to export the training block (python files) and then changing the MAX_TRAINING parameter. The easiest way is to train the model directly in the computer using those exported files.
Thanks for your quick reply. As mentioned I am 14 years old and I still go to school. The project is for a school project, so it’s 100% educational. Can you try to get approval from your university program for this, it would really help me a lot? I would also only need the increase to be temporarly till I have the new modell trained (approx. mid April 2024)
Thanks Fabi
Hi Eoin, I have another question. I would also be willing to pay for the longer training time if I can afford it from my pocket money. But I can’t even contact the sales team as I do not have an enterprise email adress but only a .gmail.com one. The contact form won’t let me submit a request with that email adress. Is there another way to contact the sales team?
Thanks, Fabi
thank you so much! That’s absolutely awesome. I’m soooooo happy. I’ll let you know how the next round of the competition goes with the larger data set.
I actually won the competition! And my image recognition was super accurate thanks to the larger data set. Thanks once again for increasing my training time. Can I keep it at that increased training time for some more months (approx. till September) because I am now quailified to participate in another national competition?
Best regards
Fabian
@Eoin and @louis Ok, Here is a short description of what I did: I created an intelligent fridge that can automatically detect food and recommend where to best store it in the fridge (in the right temperature zone so that it keeps fresh as long as possible). For the detection I used a model that I created with Edge Impulse. With my 40 min training time I was able to have 10 categories of different types of food with approx. 170 images per food type. I used 90 training cycles and after some optimization I got to an accuracy of 98,3%. And in worked really reliably when I tested it at home and when I showed it at the competition. With more training time more types of supported food would be possible. The fridge also remembers the food it contains and the shelf-live dates of the food, so when shopping you can check with a mobile app what you need to buy. And you can choose combinations of the food inside the fridge and get recipe recommendations for them. I am already thinking of how to extend the project with a camera inside the fridge to automatically detect when something is removed. I already have the camera and it takes a picture everytime the fridge door is closed. I still have to figure out how to detect if something and what was removed. So if anyone has any ideas how to best do that, that would be a great help. I am running my project on a raspberry PI 4 with a pi camera V3 and have developed everything in python.
Thanks
Fabi
@Eoin I have three temperature sensors inside the fridge. They are wired but I drilled holes through the sides of the fridge and integrated plugs in the side wall of the fridge. So they are plugable and I can transport the fridge and my pi + display + speakers independently of the fridge. And when I set everything up I just need to plug in the cables. I want to use a ready made fridge cam for inside the fridge which automatically accesses the WLAN. The vendor gave me access to the source code of the cam so I hope I can find a way to upload the images taken by the cam to some webserver and download them from there to my pi. So I plan to have only one pi and it will work indenpendently of the cam in the fridge. But I am not there yet…I still need to figure out how I can best get the pictures from the camera to the pi by adapting the camera code.
Best fabi
