Skip to main content

Video Super Resolution

gst-ai-superresolution uses an AI model to perform super-resolution reconstruction on low-resolution video, generating high-resolution output frames.

Uses the QuickSRNet Small model (~42 KB), suitable for real-time inference.

Prerequisites

Steps

1. Verify Model

radxa@airbox$
ls -l /etc/models/quicksrnetsmall_quantized.tflite

Super resolution does not require a label file.

2. View Configuration

radxa@airbox$
cat /etc/configs/config-superresolution.json

Key fields:

FieldDefaultDescription
input-file-path/etc/media/video.mp4Input video path

Note: The super resolution config uses input-file-path (not file-path).

3. Run

radxa@airbox$
gst-ai-superresolution --config-file=/etc/configs/config-superresolution.json

Press Ctrl + C to stop.

Expected Output

Terminal output:

Running app with model: /etc/models/quicksrnetsmall_quantized.tflite
VERBOSE: Replacing 10 out of 13 node(s) with delegate (TfLiteQnnDelegate) node, yielding 3 partitions for the whole graph.
Pipeline state changed from PAUSED to PLAYING

The display shows the super-resolved video output with clearer details compared to the original input.

Validation

  • Replacing 10 out of 13 node(s): 10 operators delegated to DSP, 3 fall back to CPU
  • Pipeline reaches PLAYING state
  • Display shows super-resolved video

Some operators (Reshape, Transpose) are not supported on DSP and automatically fall back to CPU. This does not affect overall functionality.

How It Works

QuickSRNet is an ultra-lightweight super-resolution network designed for mobile and embedded devices. Pipeline flow:

filesrc → qtdemux → h264parse → v4l2h264dec

qtimlvconverter

qtimltflite (DSP + CPU)

qtivcomposer

waylandsink

    You need to be logged into GitHub to post a comment. If you are already logged in, please ignore this message.

    Radxa-docs © 2026 by Radxa Computer (Shenzhen) Co.,Ltd. is licensed under CC BY 4.0