Performance Optimization for Large Image Sliced Prediction

[maxpmx]

Hi,

I'm currently working on a project using SAHI for large image segmentation (10000 x 10000 pixels) with 512 x 512 pixel patches using the get_sliced_prediction function. We've encountered a significant performance bottleneck where the processing is taking several days to complete. Our observations suggest that the merging process might not be utilizing multi-threading capabilities.

Current setup:

• Image size: 10000 x 10000 pixels
• Patch size: 512 x 512 pixels
• Using: get_sliced_prediction function

Issues:

• Very slow processing time (multiple days)
• Seems to be single-threaded during the merging process

Could you please:

• Suggest any possible optimizations or best practices to accelerate the process
• Consider adding multi-threading support for the merging process in future releases if not already available

Any guidance on improving the performance would be greatly appreciated. Thank you for your time and support!