evaluation/README.md

Rex-Omni Evaluation Guide

This guide shows how to download evaluation data, unpack images, and run Rex-Omni evaluations across datasets and task types.

1 Install FastEvaluate (required for COCO/LVIS metrics)

cd evaluation/fastevaluate
pip install -e .

2 Download datasets

• Source: https://huggingface.co/datasets/Mountchicken/Rex-Omni-EvalData
• After downloading, the directory layout should look like Rex-Omni-Eval/ with images packaged as .tar.gz files. Example on disk:

/.../Rex-Omni-Eval
  *.tar.gz               # per-dataset image archives (e.g., coco.tar.gz, hiertext.tar.gz, ...)
  _annotations/          # JSONL annotations (multiple eval types)
  _rex_omni_eval_results # The evaluation results of Rex-Omni

Unpack the image archives before running:

cd /path/to/Rex-Omni-Eval
for f in *.tar.gz; do
  echo "Extracting $f" && tar -xzf "$f"
done

3 Evaluation

The evaluation is seperated into two categories:

1. COCO/LVIS text-prompt evaluation
2. Other datasets (box/point/visual-prompt)

COCO/LVIS text-prompt evaluation in box format

For text prompt evaluation on COCO and LVIS dataset (box format), run the following script

• For COCO evaluation

bash evaluation/scrpts/eval_coco.sh \
    --model_path IDEA-Research/Rex-Omni \
    --test_jsonl Mountchicken/Rex-Omni-Eval/annotations/box_eval/COCO.jsonl \
    --image_root Mountchicken/Rex-Omni-Eval \
    --coco_json Mountchicken/Rex-Omni-Eval/coco/instances_val2017.json \
    --output_dir Mountchicken/Rex-Omni-Eval/_rex_omni_eval_results/text_prompt_eval/COCO \

• For LVIS evaluation

bash evaluation/scrpts/eval_lvis.sh \
    --model_path IDEA-Research/Rex-Omni \
    --test_jsonl Mountchicken/Rex-Omni-Eval/annotations/box_eval/LVIS.jsonl \
    --image_root Mountchicken/Rex-Omni-Eval \
    --lvis_json Mountchicken/Rex-Omni-Eval/coco/lvis_v1_val_with_filename2.json \
    --output_dir Mountchicken/Rex-Omni-Eval/_rex_omni_eval_results/text_prompt_eval/COCO \

Other datasets and task (box/point/visual-prompt)

• For text prompt task (output box)

bash evaluation/scrpts/eval_others.sh \
    --dataset Dense200 \ # choice in Dense200, DocLayNet, HierText, HumanRef, IC15, M6Doc, RefCOCOg_test, RefCOCOg_val, SROIE, TotalText, VisDrone
    --eval_type box_eval \
    --model_path IDEA-Research/Rex-Omni \
    --image_root Mountchicken/Rex-Omni-Eval \
    --output_base Mountchicken/Rex-Omni-Eval/_rex_omni_eval_results/box_eval/

• For text prompt task (output point)

bash evaluation/scrpts/eval_others.sh \
    --dataset COCO \ # choice in COCO, Dense200, HumanRef, LVIS, RefCOCOg_test, RefCOCOg_val, VisDrone
    --eval_type point_eval \
    --model_path IDEA-Research/Rex-Omni \
    --image_root Mountchicken/Rex-Omni-Eval \
    --output_base Mountchicken/Rex-Omni-Eval/_rex_omni_eval_results/point_eval/

• For visual prompt task

bash evaluation/scrpts/eval_others.sh \
    --dataset COCO \ # choice in COCO, Dense200, FSCD_test, LVIS VisDrone
    --eval_type visual_prompt_eval \
    --model_path IDEA-Research/Rex-Omni \
    --image_root Mountchicken/Rex-Omni-Eval \
    --output_base Mountchicken/Rex-Omni-Eval/_rex_omni_eval_results/visual_prompt_eval/