Feature request: shadow augmentation for training
Feature request: shadow augmentation for training: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). When training docTR models on scanned / clean document images, the trained model often struggles on real-world photos where part of the page is covered by a shadow…
The task
When training docTR models on scanned / clean document images, the trained model often struggles on real-world photos where part of the page is covered by a shadow (hand, phone, lamp, etc.). The training data rarely contains those cases, so the model never learns to be robust to them.
Part of FineEnvs/MiMo-V2.6-RL-harbor-code.