Open source computer vision API | Hacker News Reader

Open source computer vision API

openvisionapi.com

63 pointspythops14 comments

It would be nice to have a paragraph or two explaining what it does an example output. I can't imagine a lot of people are going to go through the effort of building the source code just to see what it does. Or read all of the documentation.

Based on what I managed to find in the documentation, it looks like the only thing it does is mislabel a cactus as a cat. Hopefully it's just a bad example. But I'm sure I speak for a lot of people in that I'd like to see more about what it can do before investing any time into it.

> It would be nice to have a paragraph or two explaining what it does an example output.

This can be found in two different places

- https://openvisionapi-documentation.readthedocs.io/en/latest...

- https://github.com/openvisionapi/ova-client

> it looks like the only thing it does is mislabel a cactus as a cat.

Fair criticism, This can be improved by training a new model on new large dataset.

This is s humble start, but I'm plannig adding more robust models for object detection (EfficientDet for example), also adding new capabilities for instance segmentation etc.

This is the "http-based remote procedure calls" usage of the term API, and not the "function call definitions for a programming language" usage of the term API.

(Yes, at some level they are conceptually the same thing. But it's worth pointing out due to the already-mentioned confusion with OpenCV, and for those of us that got confused in the early 2000s when a common terminology for C/C++/Java libraries suddenly started to mean "web-hosted software as a service", and still haven't fully recovered.)

The confusion with OpenCV is addressed in the comments below

I updated the website to point out that it is an HTTP based API. hope this will address the confusion with opencv

How is it different from OpenCV?

It seems this project is heavily biased towards YOLO. OpenCV has a much broader scope.

Don't think we can compare it to opencv actually. In the context of this projet, opencv can be used as a ML backend the same way I use tensorflow.

From a quick glance through the repo, this is attempting to be an open source alternative to a hosted Vision API like Google Cloud Vision API.

There's not a lot of "there" there yet, an http API layer on top of Tensorflow and YoLo essentially. But looks like a tidy start.

This is a humble start but I'm planning to add more models for object detection (efficientdet for instance), and I'm working as well to add and endpoint for segmentation

Don't think we can compare it to opencv actually. In the context of this projet, opencv can be used as a ML backend the same way I use Tensorflow.

This seems much more focused on deep learning.

OpenCV is a large and sprawling compendium of computer vision tools that features a scads and scads of classical algorithms, image processing filters, video processing tools. . . there's even a set of GUI components in there. But not much in the way of deep learning.

Love to use it but can't see Java api's unlike OpenCV so I'm out.

you're free not to use it :)

Reply on news.ycombinator.com