Choose a hi-res model

• If Yes, then continue with Step 2.
• If No, then Unstructured will not use a high-resolution object detection model when processing your files. Set the command’s --strategy option (CLI) or strategy parameter (Python/JavaScript/TypeScript) to fast. See also Choose a partitioning strategy.

If you already have your scripts or code in place and just need help in choosing a model, then skip ahead to Step 3. Otherwise:

• To have Unstructured make its best choice on your behalf about the model to use, set the command’s --strategy option (CLI) or strategy parameter (Python/JavaScript/TypeScript) to auto. You have completed this decision-maker. See also Auto partitioning strategy logic.
• To specify a specific model, set --strategy or strategy to hi_res. Then set --hi-res-model-name (CLI), hi_res_model_name (Python), or hiResModelName (JavaScript/TypeScript) to one of the models in Step 3.

• layout_v1.1.0 generally performs better than yolox at bounding box definitions and element classification. layout_v1.1.0 is a proprietary Unstructured object detection model and is used by default, as applicable, if --hi-res-model-name, hi_res_model_name, or hiResModelName is not specified.
• yolox is also provided for backwards compatibility and originally was the replacement for detectron2_onnx.
• detectron2_onnx generally underperforms the preceding models. However, it is still accessible to maintain backwards compatibility.