Skip to main content

AI Upscaling - Performance factors

The following overview provides information about the factors that influence the performance of AI upscaling.

1. General Image Processing Requirements​

Quality Preferences​

  • High naturalness preferred?
  • High quality preferred?

Speed & Resource Utilization​

  • Short processing time preferred?
  • Low resource utilization preferred?
  • High throughput preferred?

2. Individual Image Properties​

Source Image Characteristics​

  • Source image size (input dimensions)
  • Image quality (DPI) and compression level

Subject & Composition​

  • Number and size of people in the image
  • Target size of output image
    • Determines upscaling factor
    • Determines resource requirements based on input size and upscaling factor

3. Parameter Configuration​

Upscaling Factor​

  • Higher factor = Higher resource utilization

Quality vs. Speed Prioritization​

  • Choice of processing mode:
    • Fast
    • High quality

Enhancement Settings​

  • Super Resolution intensity (naturalness)
  • Facial Reconstruction intensity (quality vs. resource utilization)
  • Threshold for small face reconstruction (quality vs. resource utilization)
    • Determines the number and size of faces that will be reconstructed
  • Processing & Blending Mode (quality vs. resource utilization)

4. Hardware Resources​

GPU Specifications​

  • Number and performance of available GPU(s)

Performance Example​

Hardware: Nvidia RTX A4000 with 16GB RAM (Ada Architecture)

  • Throughput: ~6,000 to ~10,000 images per day
  • Configuration: Medium upscaling factor (2.4x) and medium input image sizes

Key Takeaways​

  1. No Universal Processing Time: Each image's processing time varies significantly based on multiple interdependent factors.

  2. Quality vs. Speed Trade-off: Higher quality settings increase processing time and resource consumption.

  3. Hardware Impact: GPU performance directly affects throughput and processing speed.

  4. Scalability Considerations: Processing large batches requires careful balance of quality settings and available hardware resources.

  5. Configuration Optimization: Proper parameter tuning based on specific requirements and hardware capabilities is essential for optimal performance.