Hello, is it possible to get free more time for job time limit? I’m only a graduating high school student trying to do a thesis with no access to much money. My project kind of needs a lot of time for processing because it has around 3.5k-4k images in its dataset (that I still have to upload). My thesis is about creating a device that would identify between biodegradable and non-biodegradable waste that would be thrown in trash bins in order to allow only the right waste in the right trash bin. If you require any proof, you can review my project. I’m willing to do anything manageable for things to be done.
Hello @Ceosine,
I’d suggest you have a first look at these recommendation to stay within the time limit: Lower compute time - Edge Impulse Documentation
What you can do is first, drastically reduce your dataset size and increase it over time to see the impact on the accuracy.
You can also use smaller image size (96x96 instead of 240x240 for example).
Then another option is to train your model locally using BYOM and import your locally-trained model back to Edge Impulse to compile it.
Best,
Louis
1 Like
I’ve tried these already and my dataset is now around 3k images but it still doesn’t work, the job processing time for FOMO still reaches over 20 minutes and reducing the number of epochs could work but it would become insignificant that it makes the model less effective.
My 2 cents: Reduce the dataset to 30 training images, deploy it and log the real-world performance of 10,000 inferences. Then repeat with 300 training images. The accuracy will increase. Then simply make an estimating statement that 3k or 30k training images would make the Model performant. A project does not need to show great results be a success.
1 Like
okay thank you, I’ll try this solution