How to Easily!! Find Similar Images in Python using Machine Learning with Deep Image Search

Подписаться 27 тыс.

Просмотров 29 тыс.

50% 1

Join this channel to get access to perks:
/ @python-programming
If you enjoy this video, please subscribe.
✅Be my Patron: / wjbmattingly
✅PayPal: www.paypal.com/cgi-bin/webscr...
DeepImageSearch: github.com/TechyNilesh/DeepIm...
Visual Studio Code C++: cltk.pythonhumanities.com
If there's a specific video you would like to see or a tutorial series, let me know in the comments and I will try and make it.
If you liked this video, check out www.PythonHumanities.com, where I have Coding Exercises, Lessons, on-site Python shells where you can experiment with code, and a text version of the material discussed here.
You can follow me at:
/ wjb_mattingly

Наука

Опубликовано:

1 май 2022

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 54

@nileshverma1964 Год назад

Thankyou for using my library. You have explained very well

@python-programming Год назад

Thanks for the comment and thanks for such an amazing library!

@carlitos5336 Год назад

Thank you for this library!

@bh_loz9578 Год назад

Hi there I have a database with over 50k photos in it and I try to find similar photos in it but it gives me only some similar one but no the same photo that is present in the database. What is the problem? And maybe here is a way to use Deep Image Search without metadata files cause photos are constantly appending to the bd and everything should be done here quick but to create a metadata of this big amount of photos takes up to 20 hours on my vps.

@AniketBangar22 Год назад

where to import the the feature extractor class from?

@prabhacar 2 года назад

Thanks for the demo!

@python-programming 2 года назад

No problem!

@GingerBeardTom Год назад

Hi, is there a way how to save the model and then recall it for another folder? Thanks

@ImStarXpL 24 дня назад

very interesting, do you think it could work with this usecase: look for n similar images that have the same set as the input image, I mean if input image has green and blue dominant and light coming from right it finds all the image with this configuration? Or it looks just for generic attributes?

@amirnoorani5017 2 года назад

Wow you are great. thanks for sharing your knowledge

@python-programming 2 года назад

Thanks! No problem!

@micjays6887 3 месяца назад

Hello and thank you very much for the video, is there a way to export the model, to use it for example in a web app where the image is sent directly and make the prediction there?

@bh_loz9578 Год назад

@nando2818 Год назад

Hi, is there a way to apply thresholding when calling the get_similar_images method?

@bayoleems Год назад

co-ask

@gokufujison 6 месяцев назад

Does ist work if the image is cropped/rotated?

@gart1609 2 года назад

I have a similar task in NLP realm. There are two vocabs of short phrases and I have to find the closest match for each of the phrase from the first in the second vocab. The closest match can contain additional words or be a complete synonym, paraphrase of the target. Very broadly speaking I'm thinking about vectorizing each vocab entry and finding the nearest neighbor or the smallest cosine distance. Will it work, or maybe there is a better approach?

@python-programming 2 года назад

Indeed it will! I have done it in production. There are several approaches from simple TFIDF to vectorizing with BERT. I would recommend the latter. It is easier to implement and faster. You can vectorize the sentences using the TopicBert library. I have been wanting to do a video on it for a while.

@gart1609 2 года назад

@@python-programming I thought about TFIDF, but I don't think it will work on rather short, 3-5 words, phrases

@python-programming 2 года назад

@@gart1609 oh yea no. TFIDF probably will fail with phrases that short but BERT would shine, especially since you expect synonym usage

@markbremerkamp6614 Год назад

@@python-programming Interesting, Please do!

@Ony_mods Год назад

Is it possible to get % similarities using this library? Couldn't find this information at a glance

@christianrazvan 4 месяца назад

How this compares to features extraction from opencv?

@botan1996 Год назад

Thanks a lot, William! This is really helpful and the setup was more than easy. Do you see any way of handling color similiarity as well? I am dealing with a collection of fabrics and it seems the code focuses more on patterns than color. E.g., a red fabric with stiched flower results in only flower fabrics returned that are all white. Having such feature would be really nice - unsure if it already exists here. Thank you again for your time and effort - means a lot!

@python-programming Год назад

Interesting question! I would recommend checking out CLIP, a multi-modal model from OpenAI. It is relatively simple to setup. I have a video on this channel called Image Search Engine. It walks through the steps for setting it up. In theory, you may be able to search for what you're describing in your collection of images and retrieve it. It may or may not work. It's worth trying though!

@pastuh 7 месяцев назад

I think "simil-photo" works great, except missing a lot of UI functions.. Would be nice to replicate..

@idensas Год назад

Wow-wow-wow! Thank you SO much! Only because of people like you, I can learn new things/skills, and make more "powerful" programs! I couldn't have done it alone. Thx ^^

@python-programming Год назад

No problem!

@shamshirgaranm3520 Год назад

nice

@lowkeysoft 2 года назад

Would you consider something like this for an imageset that's constantly being updated? Right now I'm leaning towards a dhash based database.

@python-programming 2 года назад

Yea this would work fine, but I would add a few extra functions to the classes so that you can vectorize just the new image and rerun annoy with all the images. Annoy is quick. The longest part is rerunning VGG16 over all images

@lowkeysoft 2 года назад

@@python-programming interesting. Something to think about, thanks!

@python-programming 2 года назад

No problem!

@Arthur-fq5dd Год назад

@@python-programming Hello, is it possible for you to add this to github? It would help me a lot to move forward in my project to vectorize only the new image. Because it takes too long to vectorize them all again when adding a new image.

@TechnoFilipino Год назад

@@Arthur-fq5dd I second this.

@aotrakstar Год назад

Can this library be used to search for similar images from different urls?

@python-programming Год назад

It can but you would need to do a few different steps, e. g. Downloading the photos first. I am prepping a video and library for doing this.

@fafyshosha7517 2 года назад

Sir can i use this Library to get photos of people similar to a photo of specific person ..i mean in a shape or face?

@python-programming 2 года назад

Possibly, but it would depend. I would recommend using object detection first to extract a mask of the person first and feed that to this

@fafyshosha7517 2 года назад

@@python-programming oh that is exciting.. could you make tutorial on that sir ?

@python-programming 2 года назад

@@fafyshosha7517 working on one actually with a textbook. Will not be ready until June though

@python-programming 2 года назад

Repository: github.com/wjbmattingly/deep-image-demo

@mattwebbertime 3 месяца назад

*dead link*

@deepakshashidhar7041 2 года назад

can we assign the similarity score in this

@python-programming 2 года назад

Indeed you can!

@deepakshashidhar7041 2 года назад

@@python-programming can you help me how to do so

@python-programming 2 года назад

@@deepakshashidhar7041 oh certainly. I covered it in the video. Is that what you meant by similarity score? It’s the output from DeepImageSearch. It has both the similar images and degree of similarity

@deepakshashidhar7041 2 года назад

@@python-programming i wanted values like 0.9,0.7 like that

@deepakshashidhar7041 2 года назад

scores for each image can we acheive this