Skip to content

Latest commit

 

History

History
53 lines (46 loc) · 1.88 KB

EVAL.md

File metadata and controls

53 lines (46 loc) · 1.88 KB

Evaluation

REC Benchamrks

For referring expression evaluation on RefCOCO/g/+, please download test annotations (e.g., refcoco_val.json) and run

torchrun --nnodes=1 --nproc_per_node={num_gpus} groma/eval/eval_rec.py \
    --model-name {path_to_groma_ckpts} \
    --img-prefix {path_to_coco_train_2017} \
    --ann-file {path_to_rec_test_annotation}

LVIS-Ground Benchmark

To evaluate on LVIS-Ground proposed in the paper, please download lvis_test.json and run

python groma/eval/eval_lvis.py \
    --model-name {path_to_groma_ckpts} \
    --img-prefix {path_to_coco_train_2017} \
    --ann-file {path_to_lvis_test_annotation}

Referring Benchmarks

For region captioning evalution on Visual Genome, please download vg_test.json and run

torchrun --nnodes=1 --nproc_per_node={num_gpus} groma/eval/model_vg.py \
    --model-name {path_to_groma_ckpts} \
    --img-prefix {path_to_vg_images} \
    --ann-file {path_to_vg_test_annotation} \
    --result-dir {path_to_intermediate_result_dir}
  
python groma/eval/eval_cap.py \
    --ann-file {path_to_vg_test_annotation} \
    --result-dir {path_to_intermediate_result_dir}

Similarly, for RefCOCOg, download refcocog_cap_val.json and run

torchrun --nnodes=1 --nproc_per_node={num_gpus} groma/eval/model_refcocog.py \
    --model-name {path_to_groma_ckpts} \
    --img-prefix {path_to_coco_train_2017} \
    --ann-file {path_to_refcocog_test_annotation} \
    --result-dir {path_to_intermediate_result_dir}
  
python groma/eval/eval_cap.py \
    --ann-file {path_to_refcocog_test_annotation} \
    --result-dir {path_to_intermediate_result_dir}