Lot's of interesting technical issues in here that I'm going to think about.
I wonder if an intermediate project might be to classify does the image have the language in it. Then you're dealing with yes/no output rather than the full range of possible sentences.
Which fast.ai course is this related to? And does it point you to the Aurabesh character recognition, or was that your choice?
The question I think you are asking is does it make sense to have an "aurebesh or not" classifier before doing an aurebesh letter classifier? From a deep learning perspective, I don't think it matters. The two are both straight forward to implement (and harder to generate enough data for) and one doesn't build on the work of the other. A bit like making an "animal or not" classifier doesn't really help with making a "cat or dog" classifier.
It's from week 2 of Practical Deep Learning for Coders (course.fast.ai/videos/?lesson=2). The assignment was basically, here's a bunch of ideas about how to throw stuff together, now go forth and make an image classifier. That level of hands-off is very typical for fast.ai which is in part why I love it so much. The character recognition was my choice. I'm a big Star Wars fan! But also I was looking for images to classify and the data set was already pre-generated. The sample problem that was worked through was "teddy bear, grizzly bear or black bear" other students have done "cricket or baseball". It's a fun project.
For further actions, you may consider blocking this person and/or reporting abuse
We're a place where coders share, stay up-to-date and grow their careers.
Lot's of interesting technical issues in here that I'm going to think about.
I wonder if an intermediate project might be to classify does the image have the language in it. Then you're dealing with yes/no output rather than the full range of possible sentences.
Which fast.ai course is this related to? And does it point you to the Aurabesh character recognition, or was that your choice?
The question I think you are asking is does it make sense to have an "aurebesh or not" classifier before doing an aurebesh letter classifier? From a deep learning perspective, I don't think it matters. The two are both straight forward to implement (and harder to generate enough data for) and one doesn't build on the work of the other. A bit like making an "animal or not" classifier doesn't really help with making a "cat or dog" classifier.
It's from week 2 of Practical Deep Learning for Coders (course.fast.ai/videos/?lesson=2). The assignment was basically, here's a bunch of ideas about how to throw stuff together, now go forth and make an image classifier. That level of hands-off is very typical for fast.ai which is in part why I love it so much. The character recognition was my choice. I'm a big Star Wars fan! But also I was looking for images to classify and the data set was already pre-generated. The sample problem that was worked through was "teddy bear, grizzly bear or black bear" other students have done "cricket or baseball". It's a fun project.