Bounding Boxes, Classification, Image Annotation, Line Annotation, Polygons, Semantic Segmentation
Data Management Platform for Image Annotation

How to make workload of ML engineer easier? Data Management. Introducing new TaQadam Platform for Image Annotation. Platform.taqadam.io

What do you learn after annotating 2 million images? We were excited to come up with new tricks required for the fast growing field of computer vision, and image annotation labeling and tooling service. We have upgraded the app to meet new demands. These mobile app updates have enabled us to ensure higher accuracy and precision through touch mobile annotation and collaborative team work.  But it did not end there. On the data management side, working closely with ML engineers over the year on the projects, was a very valuable experience. It helped us to learn how to better organize our own data workflow to make the cycle of data training, testing and validation easier.

What is new: Platform TaQadam

Monitoring real-live progress

See % of completion, validation and access annotation in a real-time

Enhanced quality control tools

Guide annotation process by providing the feed of comments to the teams and management

Self-managed service and API

Project Management service, continuous stream of images, active learning

Learning by doing, and introducing solutions

Complexity of Use cases, approaches, models 

 
We move into an era where computer vision solves complex real life problems. For instance security, traffic management, and object recognition in the industrial facilities. Therefore should evolve image annotation tools and practices. We ensure our labeling tool now allows multi-tool system for hybrid annotation required (segmenting, bboxes and classifier), layering attributes associated with polygons or boxes,  allowing descriptive meta data, ranking system and image-level (i.e. combining categorization or image classification) with object-level annotation. 
 
 
In this use case of waste sorting, there is a requirement to assign polygons to each type – organic, residual, bag, paper, plastic, as well we assess with a secondary attribute visibility, and provide a descriptive meta data to an image assessing efforts of household and respective suggestions. 
 

Moving beyond AI data training to validation, testing and active ML learning 

 
In the Platform we introduce query based on status of annotation, sorting, as well as selective class results download. It also allows uploading or ingesting through API call the pre-trained ML models, whether YOLO or client’s own to optimize our annotation and continue enhancing performance of existing machine learning model.  Finally, it is always about:
 

Quality Control

 
 
Everyone knows it, but in fact it means a little more than a correct label. Now we are training ML models in the complex contextual environment. Understanding the context and delivering a shared understanding among annotating team need to be in an organized workflow. In the Platform, we introduced the comment section, choice of colors, slider to allow clients to access the quality of annotation very quickly. Real-live comments would be shown to entire team, as well as in relation to a particular image in the mobile app stream.  This is a great enhancement allowing to build “managed and trained” team behind your annotation service. 
  
 
In this use case of waste sorting,  we leave comment on specific class, and allow selective validation or rejection of the images.  No more non-quality data into the data training pipeline.
 
Interested to try our new Platform? Register. Refer.
Request demo