Buflow


Labeled Training Data at Scale

Our Approach

Our proprietary model first generates labels for segmentation tasks that you submit to our platform then a human labeler makes adjustments to achieve highly precise labels that out-perform a manual labeling process.

QA: Buflow compares tasks to ground truth at random to determine accuracy. Each labeler must maintain a minimum score to be able to stay on our platform.

Fast turn around: most tasks are delivered within hours or sooner.

Labeler Score: every task includes a labeler quality score as part of its result, we use a combination of techniques including a golden set to continuously measure labeler quality.

Highly Specialized Tools: our tools are designed from the ground up with a focus on accurate training data generation.

API Callback: once you submit your tasks via our API you can request our platform to post the results back to you when ready using a callback URL.

Tile Server Support: for ultra-high resolution (4K+) images or maps Buflow can utilize a tile server that you provide instead of a single image. This can both expedite the labeling process and improve accuracy.

Annotation Tool

Responsive labeling tool that can handle large images without slowing down the browser.

T-distributed Stochastic Neighbor Embedding (TSNE) Visualization


Dig down to your ground truth to find out where to focus your labeling efforts and areas to improve.

Stop wasting resources on creating labels that have near zero impact on your model.

Quickly visualize and interact with your data to learn where to focus on in order to create a more accurate model.


Auto-Segmentation Mask and Bounding Box

Post a task to our API and get labeled results back.

...

Binary Classification

...

Segmentation

...

Point Cloud Labeling



Lidar to camera image coordinate interpolation.

Get started


Register