Cord automates the creation of your training data through our novel micro-model approach. With Cord you can turn your labeling process into an automated assembly line and create training data for your computer vision applications at unparalleled speeds.
Hey Product Hunt! We're the creators of Cord 👋
You probably have a bunch of stuff to do today, so we will be super brief.
🙋Problem
The biggest bottleneck in computer vision AI right now is not the compute power, the number of parameters in the models, or the number of intelligent data scientists, it’s the ability to create usable labeled training datasets to feed the models. This is the blocking element holding us back from building applications to solve some of the hardest problems we are facing today.
💡 Solution
That’s why we built Cord, a platform that can automate the data annotation process for computer vision use cases. We use a toolbox we call “micro models” to break up the annotation process into smaller parts and let AI do most of the work. Our platform has helped build datasets for computer vision applications that include the diagnosis of cancer, detection of people in danger in construction sites, and management of small business inventories.
🔥How does it work?
1. Define your label structure or “ontology.” This is the set of “questions you are asking the data”. Cord allows you to create arbitrarily complex label structures and dynamically adapt them throughout the course of your projects.
2. Do a handful of manual annotations. Sometimes even just 4 hand labels are enough to start “teaching” the system what you want to annotate.
3. Build a micro model. It takes less than 5 minutes to get your first model trained and running.
4. Run the model, review the predictions, and retrain. You are now in an active learning loop!
5. Repeat as necessary. Build as many micro models as you need to cover your labeling.
6. Once the labels are set, use Cord’s SDK to create a data pipeline. You now have a working dataset ready for your model! The platform’s visualization features will also help you detect biases and imbalances in your dataset through the process.
To get started come talk to us! We are offering a special deal where we are giving out 5 hours of free human annotation to the first 5 people that sign up with us through Product Hunt. Book a demo on our website quoting ‘PRODUCT HUNT’ to qualify.
We look forward to hearing your feedback and comments! 🚀
Report
Looks really cool Eric and Ulrik. Huge devellopment since I saw the first demo 🚀🚀
@eric_landau
This has been an issue for ages, even for educational purposes. You can't really learn how to do ML, because there are literally 5 datasets.
This is awesome, I know so many healthcare orgs that have been trying to clear this hurdle in their own AI application building efforts. Have you worked with any healthcare use cases yet?
@julietroseb thanks Juliet! We have worked with orgs like Stanford Medicine and King's College London. We actually published a paper in collaboration with KCL you can see here: https://www.thieme-connect.com/p...
Report
@ulsha Ahh super interesting, I'll take a read and pass this along, thanks!
Is this something we will be able to license and run on our own servers? We are quite wary of sharing our data and labels externally - we’ve had bad experiences with that…
Hi @natalap, indeed we can do on premise deployments for enterprise. Even with our cloud solution, our philosophy and terms of services is that all your data and any models you build through the platform are 100% yours. If you don't want to go for an enterprise on-prem deployment, you can connect your cloud buckets with the software -- so no need to upload anything!
Report
That’s awesome to hear! Thanks so much for your response
Report
Do you think the same approach could potentially be applied to segmentation problems as well? What about human activity recognition from video?
Hi @sj_streicher, yes indeed, we do segmentation as well as bounding box detection and classification. Human activity recognition is one of the more interesting use cases we have worked on.
Seriously great stuff! Would've loved this in my previous gig where I was managing a team of computer visions experts who were always starving for properly annotated data sets.
Labelling is definitely the bottleneck when developing niche machine vision applications, happy to see a product that focuses on a very relevant area in the field.
Encord Active
Encord Active
inBalance
Encord Active
Encord Active
Encord Active
Scrimba
Encord Active
Kitemaker
Encord Active
Encord Active
Encord Active