Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Introducing Indra, a 1M parameter OCR model: - 49 pages/s on CPU, 11x faster than PaddleOCR and 17x Tesseract - Beats them on OmniParseBench text (41.3 vs 35.8 and 19.3) - Full layout and line bounding boxes

150,472 Aufrufe • vor 1 Monat •via X (Twitter)

46 Kommentare

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Get it here: Model - Code - Blog post -

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Indra works with 60 languages and 6 scripts (Latin, Cyrillic, Greek, Devanagari, Bengali and Chinese), plus any script from a PDF's text layer, and can handle math with a special math mode.

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Indra is built for low latency on CPU: RAG ingestion, search indexing, triage before a bigger model, etc. It has limitations because it's such a small model - handwriting, complex tables, and large pages may not work. Try for these.

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

If you run pip install indra-ocr, then indra-ocr-viewer, you get an app that you can use to test speed and different settings.

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

We'll continue to improve Indra OCR, including supporting more languages and improving accuracy. Let us know if you have feedback.

Profilbild von vik
vikvor 1 Monat

woah!

Profilbild von Prithiv Sakthi
Prithiv Sakthivor 1 Monat

Wow 💨🔥

Profilbild von Louis DCK
Louis DCKvor 1 Monat

Amazing, we need faster OCR models, good job!

Profilbild von 墨无归(奶狗版)
墨无归(奶狗版)vor 1 Monat

1M params at 49 pages/s on CPU is wild. Does it handle Chinese?

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Yes, chinese works

Profilbild von CB
CBvor 1 Monat

One banger after another, damn. Great work, curious if we'd see more companies hop away from azure document intelligence to something like this.

Profilbild von Hassan Hayat 🔥
Hassan Hayat 🔥vor 1 Monat

noice

Profilbild von Pratik Sharma
Pratik Sharmavor 1 Monat

Waaao

Profilbild von Bailey Simrell
Bailey Simrellvor 1 Monat

@datalabto Wooooooowwwww awesome release LFG

Profilbild von Luca Soldaini 🎀
Luca Soldaini 🎀vor 1 Monat

🐐

Profilbild von Marko
Markovor 1 Monat

One million parameters is small enough to sit inside the app, and CPU-only at 49 pages a second means the page you scanned never has to leave the device.

Profilbild von Louis DCK
Louis DCKvor 1 Monat

Google had a built in OCR model in chrome rhat is really good. Have you tested this? I wonder how it compares.

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

We haven't benchmarked it directly, not sure if there is an easy way to do that

Profilbild von Miraj Rambhiya
Miraj Rambhiyavor 1 Monat

@VikParuchuri A table spanning multiple pages. Does it handle that too (by handle I mean subsequent pages where the table is continued have table header information in the output even though it’s not explicit in the document page)?

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

No, this does not take multi-page context into account

Profilbild von JC Denton
JC Dentonvor 1 Monat

Multimodal? Can describe images?

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

No, this is not an llm, it only does ocr

Profilbild von -‘
-‘vor 1 Monat

Works with MRZ ocrb font?

Profilbild von Dante Perea📍Dubai 🇦🇪
Dante Perea📍Dubai 🇦🇪vor 1 Monat

How far is from SOTA in benchmarks for extracting text from scientific papers? Great job @VikParuchuri

Profilbild von Omar
Omarvor 1 Monat

what's the twist? impressive latency! how does it compare to lightocr 0.8b?

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

It's only 1m params, so handwriting, etc, are not going to be as accurate as bigger models

Profilbild von Jackon J Ray
Jackon J Rayvor 1 Monat

I have a project I shall be using this for that I needed something small for it to work. This is awesome I appreciate it.

Profilbild von ಗಣ
ಗಣvor 1 Monat

Have you already released Varuna and Mitra ? How about Rudra? Savitr ?

Profilbild von Akhy Mohamed 🔻
Akhy Mohamed 🔻vor 1 Monat

fine tune recipe?

Profilbild von vini
vinivor 1 Monat

nice work guys! any tflite version? i will try to run in a esp32s3

Profilbild von Fake Umpire
Fake Umpirevor 1 Monat

Is it better than Chandra 2?

Profilbild von Mykhailo Sorochuk
Mykhailo Sorochukvor 1 Monat

49 pages per second on a one million parameter CPU model is wild performance, especially for multilingual OCR

Profilbild von Rahul Rane
Rahul Ranevor 1 Monat

This is epic. And just in time for me to benchmark on 800 documents in all asean languages

Profilbild von Reza Sayar
Reza Sayarvor 1 Monat

👀👀👀

Profilbild von Hexabl0b
Hexabl0bvor 1 Monat

The automatic mode's use of the PDF text layer is useful, but I think the headline comparison needs that distinction up front. For scanned documents, the OCR-only score in the README is the number I'd use to choose the pipeline.

Profilbild von KanoonGPT
KanoonGPTvor 1 Monat

I had a dream something like this will happen. The good thing about Tesseract is it is multi lingual. PP-OCRv6 is a reject cause it does not handle multiple languages at same time.

Profilbild von mo
movor 1 Monat

very nice thanks for sharing

Profilbild von Ganesh Kumar
Ganesh Kumarvor 1 Monat

How much different does this model provide. Previous I saw unlimited OCR it was having too much noise.

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Most ocr models at 1B parameters or more - this is 1M params. You'll see lower accuracy, but much much better speed

Profilbild von suryeah
suryeahvor 1 Monat

@VikParuchuri do u have a white paper regarding how you minimized its latency ? I am curious to learn how u optimized the run time inference. Faster than paddleocr and 17xtesseract.

Profilbild von HAIRMAX 90
HAIRMAX 90vor 1 Monat

Available for French Language ?

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

Yes, french works!

Profilbild von Isher
Ishervor 1 Monat

Last time I checked with Indic languages google still performed better for my use case will try again

Profilbild von Vik Paruchuri
Vik Paruchurivor 1 Monat

This only does hindi, since it's such a small model. Will look into other languages soon

Profilbild von archli
archlivor 1 Monat

Where can I find the list of supported languages?

Profilbild von mo
movor 1 Monat

still using tesseract in a few projects gotta change that as it seemss

Ähnliche Videos