Faster text detection + layout
Switched model architecture for the text detection and layout models:
- 30% faster on GPU
- 4x faster on CPU
- 12x faster on MPS (M series macs)
Accuracy should be about the same, or slightly better, from my benchmarks.
Switched model architecture for the text detection and layout models:
Accuracy should be about the same, or slightly better, from my benchmarks.