Skip to main content

Performance Optimization

Tips for optimizing OpenTryOn performance.

GPU Optimization​

  1. Use GPU for inference:
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
  1. Pre-load models:
net = load_cloth_segm_model(device, checkpoint_path)
# Reuse model for multiple images
  1. Batch processing:
# Process multiple images in batches
for batch in batches:
    results = process_batch(batch)

Memory Optimization​

  1. Reduce image resolution
  2. Process in smaller batches
  3. Use CPU offloading for large models

Speed Optimization​

  1. Use smaller UNet dimensions (64 vs 128)
  2. Reduce number of diffusion steps
  3. Use quantization for inference

See Troubleshooting for more tips.