Looks to be a pretty well made site. Just a note that it breaks down a little when there are no auto-complete suggestions for the input - there's no indication at all that there are no auto-complete results or that it won't work without one, and the page appears to do nothing. I discovered this by typing the site's own domain into the input box.
Fixed now. Also added a spinner gif while processing the request. Really appreciate your feedback. Feel free to let me know if you feel anything else is lacking. I am a math guy primarily who rarely ventures out of his jupyter-notebook comfort zone but thanks to this project I learned a good bit of frontend hacking. Cheers.
Great stuff. My next suggestion would be to make the logos shown in the results clickable to search for them, so that you can explore similar ones quickly.
I'm curious if you have a particular use case in mind for this, or was it just fun to build?
That's a neat idea. Thank you. I was thinking of putting search functionality in the results page too so that it will save a click and display names of the brands but this is better. Makes it more fun to explore for the users.
I implemented this with the Indian Trademarks database with the purpose of helping brands detect obvious infringements of design to begin with. It worked reasonably well but the database comprised mostly wordmarks(text images) and badly compressed/scanned images so the results were kinda underwhelming, except for the popular brands. So I figured I should acquire a better graphics dataset to validate my algorithm and went ahead with it. The results turned out to be much better than the earlier dataset and I thought why not use this to make a fun tool and ended up doing this. I'm hoping this turns out to be a useful research tool for logo designers.
Your constructive feedback has given me the idea to explore degrees-of-separation-between-two-logos feature, something I'll hopefully implement in a couple of days. Thanks again for taking time to check this out.
Very useful case for simple computer vision. I like it.
I wonder if something like this would be useful for large brands to detect if someone was using their logo on a website, or an item picture on an e-commerce site.
Thank you and yes indeed. This, coupled with rotation invariance is ideal for design/trademark/logo infringement detection scenarios. This result for example is against the Indian Trademark database https://imgur.com/gallery/SIAwl For photos, there's Tineye of course. Color similarity was trivial and took less than 10 lines of code but the shape similarity metric turned out to be quite challenging and took me months to get a reasonably working version. Most of the methods suggested in literature worked wonderfully against MPEG 7 dataset - http://www.dabi.temple.edu/~shape/MPEG7/dataset.html - (Few of them got a precision of almost 8 out of top 10 results!) but fared miserably with large datasets of real world designs.
8 comments
[ 4.0 ms ] story [ 29.7 ms ] threadCheers
I'm curious if you have a particular use case in mind for this, or was it just fun to build?
I implemented this with the Indian Trademarks database with the purpose of helping brands detect obvious infringements of design to begin with. It worked reasonably well but the database comprised mostly wordmarks(text images) and badly compressed/scanned images so the results were kinda underwhelming, except for the popular brands. So I figured I should acquire a better graphics dataset to validate my algorithm and went ahead with it. The results turned out to be much better than the earlier dataset and I thought why not use this to make a fun tool and ended up doing this. I'm hoping this turns out to be a useful research tool for logo designers.
Your constructive feedback has given me the idea to explore degrees-of-separation-between-two-logos feature, something I'll hopefully implement in a couple of days. Thanks again for taking time to check this out.
I wonder if something like this would be useful for large brands to detect if someone was using their logo on a website, or an item picture on an e-commerce site.