lesson_85_synthetic_scene_text_spotting

Lesson 85: Synthetic Scene Text Spotting

This lesson provides a toy-first, CPU-friendly scene text spotting loop:

• synthesize one text region per image with a dense spotting score map target
• decode a short fixed-length token sequence for recognition
• reuse a local dlhub.vision.scene_text_spotting family as the spotting backbone
• optimize a compact multi-task objective (detection + recognition + auxiliary token)

Run:

python -m tracks.vision.lesson_85_synthetic_scene_text_spotting.train \
  --epochs 1 \
  --max-train-batches 2 \
  --max-eval-batches 1 \
  --device cpu