Xenova

/

table-transformer-structure-recognition-v1.1-all

like 0

Hello @Xenova ,
Thank you for the good work, having an ONNX version of this model is quite nice! I found out that only specifying the longest edge does not work with latest version of transformers (4.38.1), and using either DetrImageProcessor, DetrFeatureExtractor (deprecated) or AutoImageProcessor classes for instantiation. Rolling back to the configuration from Xenova/table-transformer-structure-recognition does seem to give good results, i.e.:

"size": {
    "longest_edge": 1000,
    "shortest_edge": 800
}

What do you think?