We are doing our graduation project in photogrammetry: close range 3d scanning, you are a life saver thank you a ton. I will open source the project once we are done. Thank you for your efforts ^^
@JohnMcclaned
2 жыл бұрын
I would love to see a video cover the same thing (as well as some object detection / search in images ) using ML models (CNNs) to capture more of the meaning behind the images.
@jamesbriggs
2 жыл бұрын
That's exactly where I'm going with these; there's a *rough* list of the future video/article chapters here: www.pinecone.io/learn/image-search/ But in the meantime, I released a video covering image embeddings with CLIP a few days ago: kzitem.info/news/bejne/mm5vx3-Lj3SXlaA CLIP is cool because it's not just for images but also text
@itzdavich
Жыл бұрын
This was very helpful and well-explained thank you!
@matt495311
Жыл бұрын
Agreed!
@jamesbriggs
Жыл бұрын
glad it was helpful!
@aka0989
Жыл бұрын
you are a life savior, thanks A LOT
@chitreshkaushik6980
Жыл бұрын
Enjoyed it, thanks for these session James.
@shaheerzaman620
2 жыл бұрын
Fantastic stuff James!
@jpablosaavedrag
Жыл бұрын
Thanks James, great explanation and go through!
@ThankYouESM
10 ай бұрын
I feel the need to avoid grayscale because it seems like BOVW can also generate images maybe through the Wave Function Collapse algorithm, but... at pixel size. However... for initial training, I might need to create edge-detection grouping sliders to remove what is irrelevant to the target images, whereas all the grass, sky, plain walls, sand, etc... can automatically be removed. Most importantly... have it show the top 50 list of the patterns it found that when clicked on will factor those features to read as 100% every 16x16 of that image classification. Just a quick brainstorm.
@venkatesanr9455
2 жыл бұрын
Thanks for videos and inputs.
@jamesbriggs
2 жыл бұрын
you're welcome thanks for watching :)
@karndeepsingh
2 жыл бұрын
I just want to understand comparison between CNN features and Bag of visual Features? What are the adavantages and disadvantages over CNN features for the same image. Thanks
@jamesbriggs
2 жыл бұрын
I'll cover this in full soon, but two biggest differences are that CNNs can learn more abstract features, SIFT feature extractor we use here focuses on identifying features from edges, which is limited. The result of that is that SIFT is good at identifying images with similar edges/patterns, CNNs are better at identifying different "concepts" like "this is a dog". The downside of CNNs vs SIFT is that the results of SIFT are more interpretable, eg we know these images are being returned because of shared edges/patterns in the image, CNNs are much more complex and understanding exactly why they might place different images in a similar vector space is not always clear.
@muhammadhammadkhan1289
2 жыл бұрын
Amazing content as always, can you please tell me which theme do you use for jupyter notebook?
@jamesbriggs
2 жыл бұрын
thanks! this is "Chandrian Owl Theme"
@VauRDeC
2 жыл бұрын
thanks mate !
@jamesbriggs
2 жыл бұрын
glad it helped :)
@chifaasebbane5984
Жыл бұрын
hi good job.very usful.plz where i can found the code of the model
@chifaasebbane5984
Жыл бұрын
???????
@celestial_x
Жыл бұрын
thanks James, Really helped me. 🙏🫂
@celestial_x
Жыл бұрын
Subbed :) you deserve it!!
@jamesbriggs
Жыл бұрын
That's awesome, glad it helped and hope I help more in the future 😄
Пікірлер: 24