How this Helps
This feature enables teams to streamline repetitive or manual tasks, reducing time spent on video review and dataset cleanup. It’s particularly valuable for computer vision teams managing large-scale video datasets across multiple environments.
- [Use case 1]: [brief description]
- [Use case 2]: [brief description]
- [Use case 3]: [brief description]
[Feature Architecture / How It Works]
The diagram below outlines the key components of this feature in Visual Layer:![[Alt text for diagram]](https://mintlify.s3.us-west-1.amazonaws.com/visual-layer/img/[relevant-diagram].png)
| Component | Description | |
|---|---|---|
| 1 | Triggers | Events or user actions that initiate the workflow or feature process. |
| 2 | Steps | Sequential actions or evaluations taken on the video or image data. |
| 3 | Flow Controls | Logic that governs conditional execution of steps, such as filters or decision trees. |
| 4 | Outputs | Final outputs such as metadata exports, updated video versions, or metrics. |
You can use Visual Layer Copilot to accelerate the setup process and generate optimized flows automatically.
Example I: Auto-Remove Blurry Frames Before Annotation
Example I: Auto-Remove Blurry Frames Before Annotation
A computer vision team at ACME Robotics needs a pre-processing step that removes low-quality frames before the data is sent for labeling.This process begins with an on-demand trigger that scans uploaded datasets. The system evaluates each frame for sharpness and flags blurry frames for exclusion. Once identified, those frames are automatically removed or isolated in a separate dataset version.The result is a cleaner, more efficient annotation pipeline with higher training data quality and lower cost.
Using [Feature Name]
With [Feature Name], you can:- [action 1]
- [action 2]
- [action 3]
- [action 4]
[First sub-heading — e.g., How It Works]
The following is an overview of how the process works:1
Upload your video dataset
Upload a dataset using the Visual Layer UI or API. Once uploaded, the system will begin metadata indexing.
2
Enable frame quality checks
Enable frame quality filters from the configuration panel.
3
Define thresholds for blur detection
Set acceptable ranges for visual clarity, based on your project’s needs.
4
Review flagged frames
A preview set of flagged frames will be shown for validation. You can confirm removals or override false positives.
5
Commit clean dataset version
Once validated, save the filtered dataset as a new version for downstream tasks like labeling or fine-tuning.
Recommendations and Limitations
- Works best on datasets larger than 500 frames
- Thresholds may need adjustment for infrared or thermal footage
- Currently supports
.mp4,.avi, and.movformats - GPU-accelerated processing is only available on Enterprise plans