Hacker News new | comments | show | ask | jobs | submit login
Show HN: Bounding-box labeler tool to generate the training data for YOLO v2 (github.com)
64 points by cartucho 8 months ago | hide | past | web | favorite | 16 comments



Thanks for sharing.

We've also been working on a similar tool (although does not output to YOLO format): https://www.labelbox.io/

We've open sourced the labeling interface for anyone to easily build and support labeling any data - as long as it can be loaded in a browser. Learn more here: https://github.com/Labelbox/Labelbox

About Labelbox:

Labelbox is a enterprise grade and cloud based tool to easily label data for machine learning. Labelbox streamlines data labeling workflow, from micro labeling projects for quick R&D to production grade projects requiring hundreds of collaborators. It is agnostic to data type and has open source labeling frontend with already built templates for image classification & segmentation and text classification. One can label any other kind of datasets by creating a custom labeling interface with javascript API (labeling-api.js). Additional feature includes exporting data in JSON/CSV with auto generated image masks, project & team management and labeling analytics.


This is a much better alternative and works completely in client side js

http://www.robots.ox.ac.uk/~vgg/software/via/

http://www.robots.ox.ac.uk/~vgg/software/via/via_demo.html


The idea is to use OpenCV so that later it uses SIFT and Tracking algorithms to make labeling easier. I wanted this tool to give us automatic suggestions for the labels!


I had to look it up:

> "You only look once (YOLO) is a state-of-the-art, real-time object detection system."

https://pjreddie.com/darknet/yolo/

I'd also include the "image classifier" keyword.


You can go through a video frame-by-frame with YouTube, and it's worth doing it for the video on their landing page for moments such as: https://i.imgur.com/5tRzFsZ.jpg

https://www.youtube.com/watch?v=VOC3huqHrss


Yes! There are very cool tutorials on how to train Yolo to detect your own objects:

https://timebutt.github.io/static/how-to-train-yolov2-to-det...


It’s an object detector, not an image classifier.


Key to annotation is using pre trained or partially trained object detectors along with similarity search to quickly highlight unlabelled, similar objects in a dataset. Think of group labelling instead of labeling one object at a time. One of the open source implementations of object search out there is https://github.com/beniz/deepdetect/tree/master/demo/objsear... and that you can build your own group annotation tool around.


Interesting project, many thanks for sharing!

On a side note: I am working on a client side annotation tool which is backed by OpenCV and Mask R-CNN. It is still in the early stages (and may contain some bugs), but in case you are interested, you can check it out here [1]

[1] https://imagemonkey.io/annotate

The whole service is running on a really small instance at the moment (will be moved soon), so it's rather slow at the moment.


This is really cool! Have you tried using superpixel segmentation? It works well for some biological applications. Grabcut is also immensely useful.

We've been working on a platform for medical image and video annotation tasks. This means we need to support everything from DICOM to large pathology images, and endoscopy videos. We also have it connected to deep learning networks (e.g. Inception v3, YOLO, ENet) so you can easily train or download the JSON for offline analysis.

We'd love to have beta testers from the HN community: https://semantic.md/annotate.html


Thank you! I will give it a try with superpixel and grabcup!


If you are using Mac OS X, you can use RectLabel. An image annotation tool to label images for bounding box object detection and segmentation. https://rectlabel.com

Key features:

Drawing bounding box, polygon, and cubic bezier

1-click buttons make your labeling work faster

Customize the label dialog to combine with attributes

Settings for objects, attributes, hotkeys, and labeling fast

Layer order for overlapped boxes

Zoom in on a point

Quick zoom to existing boxes

Support the PASCAL VOC format


I've used a bunch of these for object detection labeling.. the best one i found so far is labelmg (free) and rectlabel for Pascal VOC format.


What makes this more appealing than a web interface? Presumably you still need a service to manage the data, and deploying to labelers will be difficult.


Thanks for the comment so the idea is to use OpenCV so that later it also supports video format and uses SIFT and Tracking OpenCV algorithms to make labeling easier.

I wanted it to give automatic suggestions.


Ohhh I see, cool




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: