huggingface / pixparse

Pixel Parsing. A reproduction of OCR-free end-to-end document understanding models with open data
11 stars 3 forks source link

`image_preprocess` and `anno_preprocess` not used in `hf_dataset` source #38

Open molbap opened 9 months ago

molbap commented 9 months ago

As title. image preprocess and anno preprocess are not passed in hf_dataset loader. They are still used correctly in specific tasks because of Collate classes that are specific to these tasks and invoke processing steps correctly but would be better have all preprocessing steps called in the same fashion, less error-prone.