How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
readwriteweb.com
How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
1–10 of 15 posts
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#2Spoiler: The algorithms actually work by doing text analysis on the captions and other metadata, and no actual image analysis.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#3Spoiler: The algorithms actually work by doing text analysis on the captions and other metadata, and no actual image analysis.
the algorithm needed to execute quickly. image analysis takes too long.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#4Spoiler: The algorithms actually work by doing text analysis on the captions and other metadata, and no actual image analysis.
the algorithm needed to execute quickly. image analysis takes too long.
What's the bottleneck? Bandwidth for downloading images? Understanding of an algorithm that would do the job?
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#5Earlier quoted context omitted.
the algorithm needed to execute quickly. image analysis takes too long.
What's the bottleneck? Bandwidth for downloading images? Understanding of an algorithm that would do the job?
Difficulty in creating an algorithm.
There are ways to get it done algorithmically, however, the challenge is in getting enough data. 30,000 images is too low. you would need a few million, then just simple machine learning algorithms would work.
The latest machine learning techniques such as unsupervised deep learning might work, with millions of unlabeled images and the 30,000 labelled.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#6Earlier quoted context omitted.
the algorithm needed to execute quickly. image analysis takes too long.
What's the bottleneck? Bandwidth for downloading images? Understanding of an algorithm that would do the job?
Bandwidth and general cumbersomeness of dealing with larger amounts of data with starving-startup resources. I actually spent about a decade of my career focused on image processing, and while I love it's power, I knew how much of an engineering challenge it can be at massive scale. I need to do a blog post about this, since I know my choice is a bit surprising and needs explanation.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#7Spoiler: The algorithms actually work by doing text analysis on the captions and other metadata, and no actual image analysis.
Spoiler 2: They achieved the low cost by exploiting Kaggle's army of machine learning practitioner suckers/volunteers/competitors.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#8Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#9Is 30,000 photos a large enough dataset for a ML exercise like this?
It might just be journalistic simplifying but I wonder if they overfit their data with such specific "good" and "bad" words:
Among the best words: Peru, Cambodia, Michigan, tombs, trails and boats. What photo captions are the most likely to signify a bad photo for a travel magazine? San Jose, mommy, graduation and CEO, Warden says.
Re: How Jetpac Built a Photo Quality Algorithm for $5k in 3 Weeks
#10The 60 second challenge thing strikes me as extremely short-sighted. It is impressive to be able to solve a series of tiny bugs with just 60 seconds for each one, but I don't think it's the right metric to find the elusive "10x" programmer.
It's my opinion that "10x" programmers are an order of magnitude faster than their peers not because they are lightning-fast at spitting out code (although they may be). They're faster because they can analyze the problem, drill down to its essence, and come up with an effective plan for solving it. Contrast this with a poor programmer, who might code circles around the problem, wasting time writing code that could just be taken from off the shelf. It doesn't matter if the poor programmer somehow is brilliant at punching out correct code if the code they're punching out didn't need to be written in the first place.
It's like selecting creative writers by holding a contest to see which ones can spot a grammatical mistake the most quickly. Sure, that's a helpful skill for a writer to have, but it doesn't mean that they'll write anything worth reading.