fishxinyu/flexcombine-bench/tools/Video-Depth-Anything/benchmark - Storage Bucket

Benchmark

Prepare Datasets

Download datasets from the following links: Sintel, KITTI, Bonn, ScanNet, NYUv2

pip3 install natsort
cd benchmark/dataset_extract
python3 dataset_extrtact${dataset}.py

This script will extract the dataset to the benchmark/dataset_extract/dataset folder. It will also generate the json file for the dataset.

Run inference

python3 benchmark/infer/infer.py \
    --infer_path ${out_path} \
    --json_file ${json_path} \
    --datasets ${dataset}

Options:

• --infer_path: path to save the output results
• --json_file: path to the json file for the dataset, like sintel_video.json, scannet_video_500.json, scannet_video_tae.json
• --datasets: dataset name, choose from sintel, kitti, bonn, scannet, nyuv2

Run evaluation

## tae
bash benchmark/eval/eval_tae.sh ${out_path} benchmark/dataset_extract/dataset
## ~110frame like DepthCrafter
bash benchmark/eval/eval.sh ${out_path} benchmark/dataset_extract/dataset
## ~500frame 
bash benchmark/eval/eval_500.sh ${out_path} benchmark/dataset_extract/dataset