video_backbone剖析

unitree_z1_stackbox/case4/output/sparse_8/inference/round_summary.csv

@@ -0,0 +1,13 @@
+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_stackbox-vid35-fs4,unitree_z1_stackbox,0,34.59976005554199,34.61554956436157,141.86938977241516,,,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,1,34.54426956176758,34.29725003242493,142.3462598323822,191.86758422851562,0.20026391744613647,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,2,33.964932441711426,34.302324056625366,140.32403826713562,157.78009033203125,0.30717968940734863,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,3,34.355379581451416,34.40061974525452,141.09501934051514,178.2445831298828,0.28292834758758545,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,4,34.35292649269104,33.68916344642639,139.22638654708862,187.14605712890625,0.30522143840789795,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,5,34.516475439071655,34.26928973197937,141.8010392189026,176.5004119873047,0.2606137990951538,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,6,34.27536368370056,34.247106075286865,141.1553394794464,161.79750061035156,0.2811535596847534,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,7,34.446311473846436,34.475435972213745,141.3955273628235,200.7637939453125,0.42580825090408325,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,8,34.74917650222778,34.34452247619629,143.47135639190674,185.031494140625,0.31717449426651,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,9,33.78364181518555,34.44818687438965,141.8448030948639,177.88961791992188,0.20682477951049805,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,10,34.29461979866028,34.656813621520996,143.81123065948486,132.37266540527344,0.16953390836715698,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,11,34.664243936538696,33.73895835876465,140.96735668182373,133.3111572265625,0.2785102128982544,

unitree_z1_stackbox/case4/output/sparse_8/inference/sample_summary.csv

@@ -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_stackbox-vid35-fs4,unitree_z1_stackbox,policy,34.37892506519953,,,,47.083333333333336,49.166666666666664,47.666666666666664,48.416666666666664,,171.1549960049716,0.27592839977957984,141.60897888739905,34.37892506519953,34.290434996287026,
+unitree_z1_stackbox-vid35-fs4,unitree_z1_stackbox,world_model,34.290434996287026,,,,47.416666666666664,49.333333333333336,47.666666666666664,48.416666666666664,,171.1549960049716,0.27592839977957984,141.60897888739905,34.37892506519953,34.290434996287026,

unitree_z1_stackbox/case4/run_world_model_interaction.sh

@@ -2,11 +2,11 @@ res_dir="unitree_z1_stackbox/case4"
 dataset="unitree_z1_stackbox"
 
 {
-    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 "${res_dir}/output/sparse_8" \
         --bs 1 --height 320 --width 512 \
         --unconditional_guidance_scale 1.0 \
         --ddim_steps 50 \
@@ -22,5 +22,8 @@ dataset="unitree_z1_stackbox"
         --guidance_rescale 0.7 \
         --perframe_ae \
         --analysis_log_metrics \
-        --analysis_reference_steps 50
+        --analysis_reference_steps 50 \
+        --head_schedule_steps 0 7 14 21 28 35 42 49 \
+        --head_skip_mode reuse_prediction \
+        --head_log_steps 40 43 46 47 48 49
 } 2>&1 | tee "${res_dir}/output.log"