Terrible performance when using a jpeg in combination with node.js about tesseract.js

[bijection]

Interesting! Is the raw array smaller for a png of the same size?

[rowasc]

Posting findings in case it helps someone debug this issue. It'd be great to have a couple of sets of images that are more complex (maybe) to try this out

Medium size image (960px x 720px)

Execution time from start to finish

Tesseract JPG (22Kb)

• 7836.192ms
• 7683.145ms
• 7327.576ms
• 7546.628ms

Tesseract PNG (11kb)

• 9132.705ms
• 9054.147ms
• 8672.093ms
• 9110.318ms

This is the set of images I used (dummies I created with google draw.. )

I'm running a second set of tests with bigger images (>100kb) that looks like it's going to yield pretty much the same results (>2 minutes processing time for each file on the first run, with the 177kb JPG taking ~13 second less than the 100Kb PNG)... weird.

[rowasc]

This is the "large" set of images..

[bartvanremortele]

@rowasc I can provide you with some testing images:
These have already been pre-processed. I think the images you tested with are too clean and don't have any noise so they don't apply on our use case for tesseract and tesseract.js. I will look into this issue more today and provide you with feedback when I know more.

[rowasc]

Thanks ! yes, that makes total sense.
I'll get back to this after work today.

[bartvanremortele]

@rowasc Have you looked into the performance yet? I'm interested in further contribution if I can help.

[pabx06]

bmp image width in pixel:608 height in pixel:300
TEXT is a THREE LETTTER UPERCASE sans seriff.

i get TIME spent (ms)= 5330, TEXT= OAX, CONFIDENCE 92

way too slow compared to human.
i am wondering if it has something todo with preloading of language file ?

how can i tune it up for many small stream of images ?

[jeromewu]

Please use the latest version and see if the performance issue still exists, feel free to open new issues, close this one for now.

[GrillAndChill]

@jeromewu I used the new version, but it still takes almost a minute for images with more than 50 words. I have used the below attached image. It is a .png image and less than 100kb.

So I have a POST endpoint called "/image2text".

• I have tried creating a worker and loading the lang data ahead of time (out side of the POST api call), and
• I have also tried keeping it all together inside the POST call (creating the worker and loading the lang data once the user make a POST call, then terminate the worker after the call has been fulfilled).

However, the results are the same. I didn't notice any improvement in the performance.

I used this random image from google image: