alexcg1

9 ms·

31.
by alexcg1 4y ago
Oh, nifty! This is more a demo of a PDF search engine that you could (in parts 1 thru x of the series) deploy to an intranet (for internal knowledge search) or internet (for general search), rather than a collaborative tool. For handwritten
32.
by alexcg1 4y ago
Wow, this post really took off! If anyone wants to read some of my blog posts on building PDF search engines (and the pain, torment and anguish that it causes) read: - https://medium.com/jina-ai/building-an-ai-powered-p
33.
by alexcg1 4y ago
In that case I'd use: 1. PDFSegmenter (in the notebook) - extract the images of the text (yup, it does images too) 2. An OCR Executor [0][1] from Jina Hub [2] to extract the text from the images 3. Actually splice the text chunks toget
34.
by alexcg1 4y ago
Incidentally Jina Hub [0] has a few OCR Executors [1][2] you could integrate into my notebook (though you'd have to do some rewiring to take images into account since it's a text-based notebook) [0] https://hub.jina.ai&
35.
by alexcg1 4y ago
And +1 to spaCy. I typically use it over Transformers because it's SO much faster. I just used Transformers in this example for a change. My Stack Overflow search notebook [0] uses spaCy. [0] https://colab.research.google.co
36.
by alexcg1 4y ago
If new and different phenomena means new kinds of corruption and downright weird behavior I'll end up having no hair left! Even printing the same page to PDF with Chrome and Firefox delivers quite different results. Firefox was often c
37.
by alexcg1 4y ago
I'll agree that there's quite a bit of junk punctuation in the extracted sentences (and sentence fragments), quite often from short footnotes in the Wiki articles. Getting "good" PDFs with open usage rights was a bit tri
38.
by alexcg1 4y ago
PDFSegmenter also extracts images, which can then be OCR'ed in the next step of the pipeline
39.
by alexcg1 4y ago
You mean the PDFSegmenter Executor in the notebook?
40.
by alexcg1 4y ago
We've got quite a few other notebooks for other kinds of search on the blog. Would love to hear your thoughts!
41.
by alexcg1 4y ago
The version in the notebook is just for simple text-based PDFs. I wrote some posts on our company blog[1] about the sheer agonies of dealing with PDF as a data format, so wanted to stick with as simple as possible for now. That said, I&#x27
42.
Search PDFs with Transformers and Python Notebook(colab.research.google.com)
132 points by alexcg1 4y ago|51 comments
43.
by alexcg1 4y ago
Why this tech stack specifically and not some other? There's plenty of stuff out there now
44.
by alexcg1 4y ago
So much time is taken up on student questions when they could just RTFM. Having an AI take care of that menial bullshit doesn't hurt anyone. There are higher value tasks a teacher can perform then schooling someone who didn't read
45.
Generating Graphic Novels with Deep Neural Networks – DALL-E Flow and GPT-3(betterprogramming.pub)
2 points by alexcg1 4y ago|0 comments
46.
Show HN: Neural Search for Stack Overflow(colab.research.google.com)
15 points by alexcg1 4y ago|0 comments
47.
Show HN: Build Multimodal AI Search Using Nocode(medium.com)
2 points by alexcg1 4y ago|0 comments
48.
by alexcg1 5y ago
Thanks for posting this! I feel like open source tremendously boosts development speed for loads of projects, and with our growing community and developer team at Jina, things are just getting faster and faster. At this rate we'll be r
49.
by alexcg1 5y ago
If you search like "animal food" you'll get a lot of memes related to animals and food without specifically mentioning those words (e.g. "dogs" and "eating" in the caption) so it's definitely using a
50.
by alexcg1 5y ago
Hmmm...might be a model thing. We'll look into it
51.
by alexcg1 5y ago
We rely on pre-trained models at the moment, since Jina supports loads of them out of the box. For image search we use Big Transfer Encoder ( https://github.com/jina-ai/executors/tree/main/jinahub/enc
52.
by alexcg1 5y ago
10k memes are now indexed for text. So if you search for Wonka memes you'll get a lot better results. Also swapped out the model
53.
by alexcg1 5y ago
We're using Transformers with `sentence-transformers/paraphrase-distilroberta-base-v1` model. The framework is Jina ( https://github.com/jina-ai/jina/ ) so it's pretty high-level. You can see the inde
54.
by alexcg1 5y ago
I'm indexing 10k as we speak. At least for text. Indexing 10k images will take a bit longer.
55.
by alexcg1 5y ago
You can see this link. It explains why the meme search still needs a bit of work: http://examples.jina.ai:8501/?tab=Dude,%20this%20meme%20sear...
56.
by alexcg1 5y ago
I'm putting more info here: http://examples.jina.ai:8501/?tab=Dude,%20this%20meme%20sear...
57.
by alexcg1 5y ago
Likely because: a) Only so many meme types in the full dataset b) We only indexed 1000 memes to build a toy example Had I known it would blow up I would've indexed more!
58.
by alexcg1 5y ago
That's a restriction in the front-end framework unfortunately :( Really hoping Streamlit supports that in future.
59.
by alexcg1 5y ago
Haha, that's the standard upload limit for Streamlit[1] apps. We just slapped together a quick front-end using that framework. [1] https://www.streamlit.io/
60.
by alexcg1 5y ago
Since I helped build it, I'll explain :) This example is quick and dirty, indexing only about 1,000 images from a dataset we pulled from Kaggle. So if results suck, that's due to either the meme you search for not being in the dat

More ›