search

OpenRobotLab / UniHSI

[ICLR 2024 Spotlight] Unified Human-Scene Interaction via Prompted Chain-of-Contacts
https://xizaoqu.github.io/unihsi/
152 stars 5 forks source link

Evaluate scripts. #10

Open Colmar-zlicheng opened 3 months ago

Colmar-zlicheng commented 3 months ago

Hi, can you please provide test/evaluate scripts (such as how to compute success rate and dump results)? I try to use the command below, but I met segmentation fault when I set --obj_file as sceneplan/partnet_test_simple.json.

python3 unihsi/run.py \
    --task UniHSI_PartNet\
    --test \
    --num_envs 1 \
    --cfg_env unihsi/data/cfg/humanoid_unified_interaction_scene_64.yaml \
    --cfg_train unihsi/data/cfg/train/rlg/amp_humanoid_task_deep_layer_2we.yaml \
    --motion_file motion_clips/chair_mo.npy \
    --checkpoint output/Humanoid_31-17-04-30/nn/Humanoid.pth \
    --obj_file sceneplan/partnet_train_simple.json
xizaoqu commented 3 months ago

Hi, you can evaluate it by recording the states (whether fulfilled, distance) when resetting the task and summarizing them on specified samples. I'm not sure how the segmentation fault happens.