video_backbone剖析

This commit is contained in:
qhy
2026-03-16 10:30:54 +08:00
parent 7e45eba18b
commit 8ca159d375
282 changed files with 174952 additions and 1350 deletions

View File

@@ -0,0 +1,12 @@
sample_id,scene,round_id,policy_pass_total_time_s,world_model_pass_total_time_s,round_total_time_s,latent_init_dist_to_prev_round,action_drift_vs_prev_round,psnr_full50
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,0,34.6963095664978,34.687960147857666,141.07195377349854,,,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,1,34.44554090499878,34.149985551834106,142.8440294265747,242.62477111816406,0.3331108093261719,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,2,34.16856265068054,34.74272179603577,141.69387793540955,242.17344665527344,1.0369502194225788,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,3,34.6714084148407,33.94211483001709,140.9935805797577,250.14584350585938,0.5469582676887512,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,4,35.077704668045044,35.157174825668335,142.93432450294495,254.5370635986328,0.6937128007411957,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,5,34.783668994903564,33.83684039115906,141.81956338882446,223.7366485595703,0.4414963722229004,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,6,35.127192735672,34.664276361465454,143.34680724143982,236.75621032714844,0.7516259402036667,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,7,34.36726403236389,33.87373757362366,140.53388166427612,217.3331298828125,0.5563024580478668,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,8,33.7715482711792,34.000213861465454,140.70656752586365,193.4322967529297,0.0650750994682312,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,9,34.447805404663086,34.42842245101929,142.1645531654358,197.56610107421875,0.07055008411407471,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,10,34.54317378997803,34.236499071121216,142.73940634727478,210.9322052001953,0.43225961923599243,
1 sample_id scene round_id policy_pass_total_time_s world_model_pass_total_time_s round_total_time_s latent_init_dist_to_prev_round action_drift_vs_prev_round psnr_full50
2 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 0 34.6963095664978 34.687960147857666 141.07195377349854
3 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 1 34.44554090499878 34.149985551834106 142.8440294265747 242.62477111816406 0.3331108093261719
4 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 2 34.16856265068054 34.74272179603577 141.69387793540955 242.17344665527344 1.0369502194225788
5 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 3 34.6714084148407 33.94211483001709 140.9935805797577 250.14584350585938 0.5469582676887512
6 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 4 35.077704668045044 35.157174825668335 142.93432450294495 254.5370635986328 0.6937128007411957
7 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 5 34.783668994903564 33.83684039115906 141.81956338882446 223.7366485595703 0.4414963722229004
8 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 6 35.127192735672 34.664276361465454 143.34680724143982 236.75621032714844 0.7516259402036667
9 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 7 34.36726403236389 33.87373757362366 140.53388166427612 217.3331298828125 0.5563024580478668
10 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 8 33.7715482711792 34.000213861465454 140.70656752586365 193.4322967529297 0.0650750994682312
11 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 9 34.447805404663086 34.42842245101929 142.1645531654358 197.56610107421875 0.07055008411407471
12 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 10 34.54317378997803 34.236499071121216 142.73940634727478 210.9322052001953 0.43225961923599243

View File

@@ -0,0 +1,3 @@
sample_id,scene,pass_type,pass_total_time_s,action_first_stable_step,state_first_stable_step,latent_first_stable_step,action_vs_full50_90pct_step,action_vs_full50_95pct_step,oracle_budget_action,oracle_budget_state,oracle_budget_latent,latent_init_dist_to_prev_round,action_drift_vs_prev_round,round_total_time_s,policy_pass_total_time_s,world_model_pass_total_time_s,psnr_full50
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,policy,34.554561766711146,,,,47.3,49.25,46.90909090909091,47.18181818181818,,226.92377166748048,0.49280416704714297,141.89532232284546,34.554561766711146,34.33817698738792,
unitree_z1_dual_arm_stackbox_v2-vid5-fs4,unitree_z1_dual_arm_stackbox_v2,world_model,34.33817698738792,,,,47.0,48.857142857142854,47.0,47.27272727272727,,226.92377166748048,0.49280416704714297,141.89532232284546,34.554561766711146,34.33817698738792,
1 sample_id scene pass_type pass_total_time_s action_first_stable_step state_first_stable_step latent_first_stable_step action_vs_full50_90pct_step action_vs_full50_95pct_step oracle_budget_action oracle_budget_state oracle_budget_latent latent_init_dist_to_prev_round action_drift_vs_prev_round round_total_time_s policy_pass_total_time_s world_model_pass_total_time_s psnr_full50
2 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 policy 34.554561766711146 47.3 49.25 46.90909090909091 47.18181818181818 226.92377166748048 0.49280416704714297 141.89532232284546 34.554561766711146 34.33817698738792
3 unitree_z1_dual_arm_stackbox_v2-vid5-fs4 unitree_z1_dual_arm_stackbox_v2 world_model 34.33817698738792 47.0 48.857142857142854 47.0 47.27272727272727 226.92377166748048 0.49280416704714297 141.89532232284546 34.554561766711146 34.33817698738792

File diff suppressed because it is too large Load Diff

View File

@@ -2,11 +2,11 @@ res_dir="unitree_z1_dual_arm_stackbox_v2/case1"
dataset="unitree_z1_dual_arm_stackbox_v2"
{
time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
time CUDA_VISIBLE_DEVICES=0 "${PYTHON_BIN:-python}" scripts/evaluation/world_model_interaction.py \
--seed 123 \
--ckpt_path ckpts/unifolm_wma_dual.ckpt \
--config configs/inference/world_model_interaction.yaml \
--savedir "${res_dir}/output" \
--savedir "${res_dir}/output/sparse_8" \
--bs 1 --height 320 --width 512 \
--unconditional_guidance_scale 1.0 \
--ddim_steps 50 \
@@ -22,5 +22,8 @@ dataset="unitree_z1_dual_arm_stackbox_v2"
--guidance_rescale 0.7 \
--perframe_ae \
--analysis_log_metrics \
--analysis_reference_steps 50
--analysis_reference_steps 50 \
--head_schedule_steps 0 7 14 21 28 35 42 49 \
--head_skip_mode reuse_prediction \
--head_log_steps 40 43 46 47 48 49
} 2>&1 | tee "${res_dir}/output.log"