We’ve been exploring the use of a custom model within OCI Document Understanding for processing our invoice templates. While the initial setup and header-level labelling work as expected, we’ve encountered a limitation with line-level details — particularly when the invoice contains tabular structures.
Currently, the model groups all repeated fields (e.g., unit prices or item descriptions from multiple lines) under a single label. For example, if an invoice has two lines with unit prices, both values are mapped to the same “unit price” label, making it difficult to preserve the line-wise distinction during extraction.
Since some of our invoice templates miss key fields when processed through the standard model, a custom model is essential. However, this line-level labelling challenge is a blocker for achieving accurate results.
We would appreciate your guidance on how best to approach this — particularly any recommended practices or configurations that could help the custom model handle line-level extraction more effectively.