Upload Videos

Videos enable temporal analysis and frame-by-frame annotation in Datature Vi. This guide covers everything you need to know about uploading videos, including supported formats, processing details, and Data Row considerations.

📖

Main upload guide

For general information about uploading assets (web interface, SDK, tracking progress), see Upload Assets.

---

Supported video formats

Datature Vi supports a wide range of video formats:

💡

Recommended formats

For best compatibility and performance, use MP4 (H.264) or MOV formats. These provide wide browser support and efficient processing.

---

File size and specifications

Size limits

• Maximum file size: 512 GB per video
• Recommended size: Under 100 MB for optimal upload and processing performance
• For longer videos, consider splitting into shorter segments

Video specifications

• Resolution: Any resolution supported, but videos are optimized for playback
• Frame rates: All standard frame rates supported (e.g., 15 FPS, 24 FPS, 30 FPS, 60 FPS)

---

Video processing

When you upload a video to Datature Vi, it undergoes processing to optimize it for annotation:

Processing details

Frame-by-frame processing:

• Video files are processed frame-by-frame for annotation
• Processing time depends on video length and resolution
• You can continue working while videos process in the background

Optimization for annotator:

• Videos are resized so that the longest dimension is 1024 pixels
• Some lossy compression may be applied to individual video frames
• This ensures optimal performance in the web-based annotator

Audio removal:

• Any audio tracks present in the video are removed
• Videos will play without sound on the platform
• This reduces file size and focuses on visual annotation

⏱️

Processing time

Video processing time varies based on:

• Video length (longer videos take more time)
• Resolution (higher resolution takes more time)
• Frame rate (higher FPS takes more time)

You'll receive a notification when processing is complete.

---

Data Row consumption

Understanding Data Row consumption is crucial when working with video datasets.

How Data Rows are calculated

• One video frame = One image in terms of Data Row calculation
• Each frame in a video consumes 5 Data Rows (same as a single image)
• Frame count = Video duration (seconds) × Frame rate (FPS)

Calculation examples

Example 1: Short video

• 10-second video at 30 FPS
• Frame count: 10 seconds × 30 FPS = 300 frames
• Data Rows consumed: 300 frames × 5 = 1,500 Data Rows

Example 2: One-minute video

• 60-second video at 30 FPS
• Frame count: 60 seconds × 30 FPS = 1,800 frames
• Data Rows consumed: 1,800 frames × 5 = 9,000 Data Rows

Example 3: Lower frame rate

• 10-second video at 15 FPS
• Frame count: 10 seconds × 15 FPS = 150 frames
• Data Rows consumed: 150 frames × 5 = 750 Data Rows (50% less than 30 FPS)

📊

Optimizing Data Row usage

To optimize your Data Row consumption:

• Consider using lower frame rates (e.g., 15 FPS instead of 30 FPS) if appropriate for your use case
• Split longer videos into shorter segments
• Use frame sampling if you don't need to annotate every frame

Learn more about Data Row calculation and quotas.

---

Best practices

Video preparation

File optimization:

• Use compressed video formats (MP4, WebM) to reduce file size
• Consider video resolution based on your annotation needs
• Ensure videos are properly encoded before upload
• Remove unnecessary audio tracks before upload to reduce file size

Frame rate selection:

• Use consistent frame rates within a dataset when possible
• Choose frame rates appropriate for your use case:
  • 15 FPS: Suitable for most object detection tasks, reduces Data Row consumption
  • 24 FPS: Standard for cinematic content
  • 30 FPS: Good balance for general-purpose annotation
  • 60 FPS: High temporal resolution for fast-moving objects (higher Data Row cost)

File organization:

• Use unique filenames to avoid accidentally overwriting existing assets
• Organize videos in folders by category for easier batch uploads
• Remove corrupt or improperly encoded videos before uploading

Upload strategy

Batch uploads:

• Upload videos one at a time or in small batches for better reliability
• For large video datasets, consider using the Vi SDK for programmatic uploads
• Monitor processing status before uploading additional videos

Quality checks:

• Verify video plays correctly before upload
• Ensure proper encoding and codec compatibility
• Check that video contains the content you expect to annotate
• Confirm frame rate and resolution meet your requirements

Duplicate handling:

• Be aware that uploading a file with the same name will replace the existing asset
• Any annotations linked to the replaced asset will be removed
• Use unique filenames or the SDK's duplicate handling options to prevent data loss

Planning for Data Rows

Before uploading videos:

• Calculate expected Data Row consumption based on video duration and frame rate
• Plan your dataset size according to your Data Row quota
• Consider frame rate trade-offs between annotation detail and quota usage
• Review your organization's Data Row limits

---

Video-specific troubleshooting

Video is taking a long time to process

Expected behavior:

Video processing time depends on length, resolution, and frame rate. Longer or higher-resolution videos naturally take more time.

What to do:

• Wait for processing to complete (you'll receive a notification)
• Continue working on other tasks while processing happens in the background
• For very long videos, consider splitting them into shorter segments before upload

Video quality looks degraded after upload

Expected behavior:

Videos are optimized for the annotator by resizing to 1024 pixels on the longest dimension and applying some compression.

What to do:

• This is normal and ensures optimal performance in the web annotator
• The quality is sufficient for annotation tasks
• If you need higher quality for specific use cases, contact support

Video has no sound

Expected behavior:

Audio tracks are intentionally removed from videos as Datature Vi focuses on visual annotation.

What to do:

• This is the intended behavior
• The platform is designed for computer vision tasks, not audio analysis
• If you need audio context, keep original videos separately for reference

Video file too large to upload

Possible causes:

• Video exceeds 512 GB limit (very rare)
• Video is very long or high resolution
• Video has high bitrate

Solutions:

• Compress the video using video encoding tools (e.g., FFmpeg, HandBrake)
• Reduce video resolution if appropriate for your use case
• Split long videos into shorter segments
• Lower the bitrate while maintaining acceptable quality
• Consider reducing frame rate if high temporal resolution isn't critical

Not all video frames are showing

Possible causes:

• Video is still processing
• Browser cache or display issues

Solutions:

• Wait for video processing to complete
• Refresh the page or dataset explorer
• Check the notification icon for processing status
• If issue persists after processing completes, try re-uploading the video

---

Common questions

How long does video processing take?

Processing time varies based on video length, resolution, and frame rate. Short videos (under 1 minute at 30 FPS) typically process within a few minutes. Longer videos may take proportionally longer. You'll receive a notification when processing is complete.

Can I annotate videos before processing completes?

No, you need to wait for video processing to complete before you can annotate. The video must be converted to individual frames first. You can work on other tasks while processing happens in the background.

Does frame rate affect annotation quality?

Frame rate affects temporal resolution. Higher frame rates (e.g., 60 FPS) capture more frames per second, which can be helpful for fast-moving objects but increases Data Row consumption. Lower frame rates (e.g., 15 FPS) are often sufficient for many object detection tasks and reduce quota usage.

Can I re-encode videos after upload?

No, once a video is uploaded and processed, it cannot be re-encoded. If you need different settings, you'll need to delete the video and upload a new version with the desired encoding.

What happens to variable frame rate videos?

Videos with variable frame rates (VFR) are supported, but they will be converted to a constant frame rate during processing. The platform will use the average frame rate for Data Row calculation.

Can I annotate specific frames instead of the entire video?

Yes, you can use frame sampling or selective annotation in the annotator. You don't need to annotate every frame. This can help optimize your workflow and Data Row usage.

---

Next steps

After uploading videos, you can:

• Upload annotations — Import existing video annotations if you have them
• Annotate video data — Add labels frame-by-frame using the visual annotator
• Train a model — Start training once you have annotations
• View dataset insights — Analyze your dataset statistics

---

Related resources

• Upload Assets — Main guide for uploading assets
• Upload Images — Image-specific upload guide
• Vi SDK Assets API — Complete SDK reference for asset operations
• Resource Usage — Understanding Data Rows and quotas
• Create a dataset — Learn how to create datasets