Add MOVi-C checkpoints

movi_c/dynamics/dynloc_isee_result.json

@@ -0,0 +1,341 @@
+{
+  "arm": "t30k_dynloc_adaln_mlp2_nosal",
+  "arm_type": "ocvp_par",
+  "params": 1092354,
+  "seed": 0,
+  "history": 14,
+  "rollout_train": 10,
+  "rollout_eval": 10,
+  "layers": 4,
+  "d_model": 128,
+  "ffn": 256,
+  "heads": 4,
+  "grad_clip": 0.0,
+  "batch": 128,
+  "slot_cache": "/scratch/pramish_paudel/dynloc/cache/movi_c_fix_ours_nosal_loc_v1",
+  "slot_cache_ckpt_step": 97500,
+  "residual": true,
+  "is_null": false,
+  "lr": 0.0002,
+  "schedule": "cosine",
+  "best_ckpt_step": 2500,
+  "ckpt_mismatch": true,
+  "steps": 30000,
+  "val_mse_per_step": [
+    0.008110102266073227,
+    0.007601574994623661,
+    0.006965221371501684,
+    0.0067300996743142605,
+    0.006269211880862713,
+    0.00579009810462594,
+    0.005589411128312349,
+    0.004990173038095236,
+    0.0051663899794220924,
+    0.0049730949103832245
+  ],
+  "val_mse_mean": 0.00621853768825531,
+  "val_n": 250,
+  "best_val_mse": 0.02455803607863021,
+  "val_mse_per_step_best_ckpt": [
+    0.008512862958014011,
+    0.00796794518828392,
+    0.007277728524059057,
+    0.006948260590434074,
+    0.006508857477456331,
+    0.00594237120822072,
+    0.005690095480531454,
+    0.005157830659300089,
+    0.005178199149668217,
+    0.0048654815182089806
+  ],
+  "val_mse_mean_best_ckpt": 0.006404963321983814,
+  "reported_checkpoint": "latest.pt (matches sc_dyn_eval.py)",
+  "history_curve": [
+    {
+      "step": 2500,
+      "val_mse_mean": 0.006404963321983814,
+      "val_sel_mean": 0.02455803607863021,
+      "val_mse_per_step": [
+        0.008512862958014011,
+        0.00796794518828392,
+        0.007277728524059057,
+        0.006948260590434074,
+        0.006508857477456331,
+        0.00594237120822072,
+        0.005690095480531454,
+        0.005157830659300089,
+        0.005178199149668217,
+        0.0048654815182089806
+      ],
+      "improved": true,
+      "mse_return": 0.02455803607863021,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.013665779480211906,
+      "n_hidden": 9494
+    },
+    {
+      "step": 5000,
+      "val_mse_mean": 0.006306357681751251,
+      "val_sel_mean": 0.02525232842253668,
+      "val_mse_per_step": [
+        0.008344294503331184,
+        0.007807439658790827,
+        0.007138211280107498,
+        0.006818367633968592,
+        0.006396982353180647,
+        0.005848025903105736,
+        0.005636259913444519,
+        0.0050673289224505424,
+        0.005143073387444019,
+        0.0048635960556566715
+      ],
+      "improved": false,
+      "mse_return": 0.02525232842253668,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.017290313645164566,
+      "n_hidden": 9494
+    },
+    {
+      "step": 7500,
+      "val_mse_mean": 0.006238429341465235,
+      "val_sel_mean": 0.02509479653137634,
+      "val_mse_per_step": [
+        0.008248490281403065,
+        0.0077314674854278564,
+        0.007059699855744839,
+        0.006770737003535032,
+        0.0063407220877707005,
+        0.005791786592453718,
+        0.005561412777751684,
+        0.005005727056413889,
+        0.005073198117315769,
+        0.004801054950803518
+      ],
+      "improved": false,
+      "mse_return": 0.02509479653137634,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.018293721866225954,
+      "n_hidden": 9494
+    },
+    {
+      "step": 10000,
+      "val_mse_mean": 0.00621294230222702,
+      "val_sel_mean": 0.02568126634787057,
+      "val_mse_per_step": [
+        0.008193617686629295,
+        0.007679617498070002,
+        0.007008794695138931,
+        0.00673131737858057,
+        0.006291826721280813,
+        0.005775188095867634,
+        0.005559488199651241,
+        0.004993095062673092,
+        0.005085253622382879,
+        0.004811227321624756
+      ],
+      "improved": false,
+      "mse_return": 0.02568126634787057,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.021341300347189365,
+      "n_hidden": 9494
+    },
+    {
+      "step": 12500,
+      "val_mse_mean": 0.006212425418198109,
+      "val_sel_mean": 0.026542568328119718,
+      "val_mse_per_step": [
+        0.008164948783814907,
+        0.007653121370822191,
+        0.006994482595473528,
+        0.006728359963744879,
+        0.00628137681633234,
+        0.005768297705799341,
+        0.005560820922255516,
+        0.004998740274459124,
+        0.005116244778037071,
+        0.004857858642935753
+      ],
+      "improved": false,
+      "mse_return": 0.026542568328119718,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.024497612365954092,
+      "n_hidden": 9494
+    },
+    {
+      "step": 15000,
+      "val_mse_mean": 0.006233683787286282,
+      "val_sel_mean": 0.027843263616392024,
+      "val_mse_per_step": [
+        0.008139315992593765,
+        0.007633506786078215,
+        0.006981699261814356,
+        0.006737918127328157,
+        0.006291523575782776,
+        0.005804476328194141,
+        0.005606034304946661,
+        0.005024647805839777,
+        0.005190439522266388,
+        0.004927276633679867
+      ],
+      "improved": false,
+      "mse_return": 0.027843263616392024,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.028001878897918258,
+      "n_hidden": 9494
+    },
+    {
+      "step": 17500,
+      "val_mse_mean": 0.006206076592206955,
+      "val_sel_mean": 0.027334006991398548,
+      "val_mse_per_step": [
+        0.008123431354761124,
+        0.007612705230712891,
+        0.006962866056710482,
+        0.006727399304509163,
+        0.006263909861445427,
+        0.005774762947112322,
+        0.005576295778155327,
+        0.004980000667273998,
+        0.005143681075423956,
+        0.004895718302577734
+      ],
+      "improved": false,
+      "mse_return": 0.027334006991398548,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.026185779840739772,
+      "n_hidden": 9494
+    },
+    {
+      "step": 20000,
+      "val_mse_mean": 0.006202335469424725,
+      "val_sel_mean": 0.027407146592176596,
+      "val_mse_per_step": [
+        0.008117361925542355,
+        0.007603753358125687,
+        0.006962186191231012,
+        0.006718897260725498,
+        0.006259453482925892,
+        0.005772146862000227,
+        0.0055710142478346825,
+        0.004980872385203838,
+        0.005140716675668955,
+        0.004896947648376226
+      ],
+      "improved": false,
+      "mse_return": 0.027407146592176596,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.026759298922062823,
+      "n_hidden": 9494
+    },
+    {
+      "step": 22500,
+      "val_mse_mean": 0.006225568242371082,
+      "val_sel_mean": 0.028261125542735327,
+      "val_mse_per_step": [
+        0.008115013130009174,
+        0.00760388420894742,
+        0.006971630267798901,
+        0.00673705292865634,
+        0.006277970038354397,
+        0.005794059485197067,
+        0.005596905015408993,
+        0.005000453442335129,
+        0.005181505344808102,
+        0.004977208096534014
+      ],
+      "improved": false,
+      "mse_return": 0.028261125542735327,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.029252372140605148,
+      "n_hidden": 9494
+    },
+    {
+      "step": 25000,
+      "val_mse_mean": 0.006213598418980837,
+      "val_sel_mean": 0.027970619784056686,
+      "val_mse_per_step": [
+        0.008110182359814644,
+        0.007600506767630577,
+        0.0069633531384170055,
+        0.006728691980242729,
+        0.006268386263400316,
+        0.005784665234386921,
+        0.005582557525485754,
+        0.004982630256563425,
+        0.005157871637493372,
+        0.004957135301083326
+      ],
+      "improved": false,
+      "mse_return": 0.027970619784056686,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.0281542928087754,
+      "n_hidden": 9494
+    },
+    {
+      "step": 27500,
+      "val_mse_mean": 0.0062171886675059795,
+      "val_sel_mean": 0.028122864910057787,
+      "val_mse_per_step": [
+        0.00810962077230215,
+        0.007601764984428883,
+        0.0069648693315684795,
+        0.006728035863488913,
+        0.006267393473535776,
+        0.00578977819532156,
+        0.005586795974522829,
+        0.00498813483864069,
+        0.005165279842913151,
+        0.0049702138639986515
+      ],
+      "improved": false,
+      "mse_return": 0.028122864910057787,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.028571468166032288,
+      "n_hidden": 9494
+    },
+    {
+      "step": 30000,
+      "val_mse_mean": 0.00621853768825531,
+      "val_sel_mean": 0.028171923627683528,
+      "val_mse_per_step": [
+        0.008110102266073227,
+        0.007601574994623661,
+        0.006965221371501684,
+        0.0067300996743142605,
+        0.006269211880862713,
+        0.00579009810462594,
+        0.005589411128312349,
+        0.004990173038095236,
+        0.0051663899794220924,
+        0.0049730949103832245
+      ],
+      "improved": false,
+      "mse_return": 0.028171923627683528,
+      "n_return": 393,
+      "mse_hidden_vs_held": 0.028695666014030653,
+      "n_hidden": 9494
+    }
+  ],
+  "stop_reason": "step budget reached",
+  "wallclock_s": 16526.6,
+  "supervise": "dense",
+  "vis_mask": true,
+  "vis_mask_eval": {
+    "mse_return": 0.02455803607863021,
+    "n_return": 393,
+    "mse_return_app60": 0.02571689400054116,
+    "mse_return_pose4": 0.007175191402738634,
+    "mse_hidden_vs_held": 0.013665779480211906,
+    "n_hidden": 9494,
+    "definition": "masked mode: curve = per-step MSE over slot-frames whose target is VISIBLE by the evidence flag; mse_return = MSE on targets visible now and hidden one frame before (the first frame after a hide); mse_hidden_vs_held = the prediction against the HELD encoder state on hidden slot-frames -- a diagnostic, not a target"
+  },
+  "vis_mask_actions": {
+    "lo": 60,
+    "hi": 64,
+    "rule": "both",
+    "hidden": "token",
+    "dropout": 0.5,
+    "mode": "sequence"
+  },
+  "selected_on": "mse_return (first visible frame after a hide)"
+}

movi_c/dynamics/dynloc_isee_step30000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8def6b40819e35d543b41027721ae95dc214e0534f44f80af419e22cf61e54f3
+size 4400437

movi_c/dynamics/dynloc_slotcontrast_result.json

@@ -0,0 +1,341 @@
+{
+  "arm": "t30k_dynloc_adaln_mlp2full_sc",
+  "arm_type": "ocvp_par",
+  "params": 1096194,
+  "seed": 0,
+  "history": 14,
+  "rollout_train": 10,
+  "rollout_eval": 10,
+  "layers": 4,
+  "d_model": 128,
+  "ffn": 256,
+  "heads": 4,
+  "grad_clip": 0.0,
+  "batch": 128,
+  "slot_cache": "/scratch/pramish_paudel/dynloc/cache/movi_c_sc_loc_v1",
+  "slot_cache_ckpt_step": 75000,
+  "residual": true,
+  "is_null": false,
+  "lr": 0.0002,
+  "schedule": "cosine",
+  "best_ckpt_step": 2500,
+  "ckpt_mismatch": true,
+  "steps": 30000,
+  "val_mse_per_step": [
+    0.004230174235999584,
+    0.0039709582924842834,
+    0.0034490134567022324,
+    0.0034540535416454077,
+    0.00334492907859385,
+    0.002901249099522829,
+    0.002730236854404211,
+    0.0023543424904346466,
+    0.002480764640495181,
+    0.002488482976332307
+  ],
+  "val_mse_mean": 0.0031404204200953245,
+  "val_n": 250,
+  "best_val_mse": 0.018236989051487313,
+  "val_mse_per_step_best_ckpt": [
+    0.004392361734062433,
+    0.004150992725044489,
+    0.0035815390292555094,
+    0.003558580530807376,
+    0.003347693709656596,
+    0.0029854848980903625,
+    0.002794310450553894,
+    0.0024524929467588663,
+    0.0024285425897687674,
+    0.0024512028321623802
+  ],
+  "val_mse_mean_best_ckpt": 0.0032143201678991318,
+  "reported_checkpoint": "latest.pt (matches sc_dyn_eval.py)",
+  "history_curve": [
+    {
+      "step": 2500,
+      "val_mse_mean": 0.0032143201678991318,
+      "val_sel_mean": 0.018236989051487313,
+      "val_mse_per_step": [
+        0.004392361734062433,
+        0.004150992725044489,
+        0.0035815390292555094,
+        0.003558580530807376,
+        0.003347693709656596,
+        0.0029854848980903625,
+        0.002794310450553894,
+        0.0024524929467588663,
+        0.0024285425897687674,
+        0.0024512028321623802
+      ],
+      "improved": true,
+      "mse_return": 0.018236989051487313,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.014227691593869373,
+      "n_hidden": 7829
+    },
+    {
+      "step": 5000,
+      "val_mse_mean": 0.003117509186267853,
+      "val_sel_mean": 0.01896569104025804,
+      "val_mse_per_step": [
+        0.004260714165866375,
+        0.004004920832812786,
+        0.0034539673943072557,
+        0.0034332044888287783,
+        0.0032494175247848034,
+        0.002902509644627571,
+        0.002700238022953272,
+        0.002368729328736663,
+        0.0023890668526291847,
+        0.002412323374301195
+      ],
+      "improved": false,
+      "mse_return": 0.01896569104025804,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.01960132222231023,
+      "n_hidden": 7829
+    },
+    {
+      "step": 7500,
+      "val_mse_mean": 0.0030946366023272276,
+      "val_sel_mean": 0.019621311083372058,
+      "val_mse_per_step": [
+        0.004224932752549648,
+        0.003946831915527582,
+        0.003412453457713127,
+        0.0034146925900131464,
+        0.0032369340769946575,
+        0.002895557088777423,
+        0.002682217862457037,
+        0.002343174070119858,
+        0.0023747654631733894,
+        0.002414807677268982
+      ],
+      "improved": false,
+      "mse_return": 0.019621311083372058,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.022037287083663676,
+      "n_hidden": 7829
+    },
+    {
+      "step": 10000,
+      "val_mse_mean": 0.003112945705652237,
+      "val_sel_mean": 0.021313152024162684,
+      "val_mse_per_step": [
+        0.00422363867983222,
+        0.003937274683266878,
+        0.003418315900489688,
+        0.0034197522327303886,
+        0.0032832305878400803,
+        0.0029023350216448307,
+        0.002710769884288311,
+        0.002345699816942215,
+        0.002441387390717864,
+        0.0024470521602779627
+      ],
+      "improved": false,
+      "mse_return": 0.021313152024162684,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.025459208423294376,
+      "n_hidden": 7829
+    },
+    {
+      "step": 12500,
+      "val_mse_mean": 0.003118503140285611,
+      "val_sel_mean": 0.021690625201734863,
+      "val_mse_per_step": [
+        0.004221350885927677,
+        0.0039450922049582005,
+        0.003432825906202197,
+        0.00342581607401371,
+        0.003298505675047636,
+        0.002910003764554858,
+        0.002704655984416604,
+        0.0023522102274000645,
+        0.00245903292670846,
+        0.0024355361238121986
+      ],
+      "improved": false,
+      "mse_return": 0.021690625201734863,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.027526956573575984,
+      "n_hidden": 7829
+    },
+    {
+      "step": 15000,
+      "val_mse_mean": 0.0031178249046206474,
+      "val_sel_mean": 0.021805687012078936,
+      "val_mse_per_step": [
+        0.004224013537168503,
+        0.003951601684093475,
+        0.00343530741520226,
+        0.0034389286302030087,
+        0.0032990011386573315,
+        0.0028980006463825703,
+        0.002705998718738556,
+        0.0023428425192832947,
+        0.00244252011179924,
+        0.0024400341790169477
+      ],
+      "improved": false,
+      "mse_return": 0.021805687012078936,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.024425479305675166,
+      "n_hidden": 7829
+    },
+    {
+      "step": 17500,
+      "val_mse_mean": 0.003139480948448181,
+      "val_sel_mean": 0.022978871549147906,
+      "val_mse_per_step": [
+        0.004228271543979645,
+        0.003966455813497305,
+        0.0034530425909906626,
+        0.0034518162719905376,
+        0.00334844458848238,
+        0.0029128161258995533,
+        0.0027285118121653795,
+        0.002355186967179179,
+        0.0024791890755295753,
+        0.0024710737634450197
+      ],
+      "improved": false,
+      "mse_return": 0.022978871549147906,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.02882077018181351,
+      "n_hidden": 7829
+    },
+    {
+      "step": 20000,
+      "val_mse_mean": 0.0031338303815573454,
+      "val_sel_mean": 0.02278164905284096,
+      "val_mse_per_step": [
+        0.004229405429214239,
+        0.003966412972658873,
+        0.003443368012085557,
+        0.003447669791057706,
+        0.0033463563304394484,
+        0.0028968898113816977,
+        0.002726573497056961,
+        0.0023487911093980074,
+        0.002460930962115526,
+        0.002471907064318657
+      ],
+      "improved": false,
+      "mse_return": 0.02278164905284096,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.026375393313852392,
+      "n_hidden": 7829
+    },
+    {
+      "step": 22500,
+      "val_mse_mean": 0.0031413782853633165,
+      "val_sel_mean": 0.023208157187367715,
+      "val_mse_per_step": [
+        0.004231938626617193,
+        0.003971157129853964,
+        0.00344888586550951,
+        0.003454843070358038,
+        0.003351630410179496,
+        0.002904782071709633,
+        0.002732868306338787,
+        0.002353195333853364,
+        0.0024792319163680077,
+        0.0024852484930306673
+      ],
+      "improved": false,
+      "mse_return": 0.023208157187367715,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.027290345943663856,
+      "n_hidden": 7829
+    },
+    {
+      "step": 25000,
+      "val_mse_mean": 0.0031325598247349262,
+      "val_sel_mean": 0.022828003495547904,
+      "val_mse_per_step": [
+        0.004225472453981638,
+        0.003966868855059147,
+        0.0034443391487002373,
+        0.0034460704773664474,
+        0.0033320412039756775,
+        0.002895192475989461,
+        0.0027222102507948875,
+        0.002350135939195752,
+        0.002466977806761861,
+        0.0024762891698628664
+      ],
+      "improved": false,
+      "mse_return": 0.022828003495547904,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.026189728553061557,
+      "n_hidden": 7829
+    },
+    {
+      "step": 27500,
+      "val_mse_mean": 0.003138574305921793,
+      "val_sel_mean": 0.023134391195272684,
+      "val_mse_per_step": [
+        0.004228797275573015,
+        0.0039690290577709675,
+        0.0034475072752684355,
+        0.0034531797282397747,
+        0.003342245938256383,
+        0.002899585058912635,
+        0.0027286368422210217,
+        0.0023530523758381605,
+        0.0024786123540252447,
+        0.002485095988959074
+      ],
+      "improved": false,
+      "mse_return": 0.023134391195272684,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.027375193600880567,
+      "n_hidden": 7829
+    },
+    {
+      "step": 30000,
+      "val_mse_mean": 0.0031404204200953245,
+      "val_sel_mean": 0.023201204452647672,
+      "val_mse_per_step": [
+        0.004230174235999584,
+        0.0039709582924842834,
+        0.0034490134567022324,
+        0.0034540535416454077,
+        0.00334492907859385,
+        0.002901249099522829,
+        0.002730236854404211,
+        0.0023543424904346466,
+        0.002480764640495181,
+        0.002488482976332307
+      ],
+      "improved": false,
+      "mse_return": 0.023201204452647672,
+      "n_return": 233,
+      "mse_hidden_vs_held": 0.02746752545148085,
+      "n_hidden": 7829
+    }
+  ],
+  "stop_reason": "step budget reached",
+  "wallclock_s": 16472.9,
+  "supervise": "dense",
+  "vis_mask": true,
+  "vis_mask_eval": {
+    "mse_return": 0.018236989051487313,
+    "n_return": 233,
+    "mse_return_app60": 0.01828306395352654,
+    "mse_return_pose4": 0.017545878983106736,
+    "mse_hidden_vs_held": 0.014227691593869373,
+    "n_hidden": 7829,
+    "definition": "masked mode: curve = per-step MSE over slot-frames whose target is VISIBLE by the evidence flag; mse_return = MSE on targets visible now and hidden one frame before (the first frame after a hide); mse_hidden_vs_held = the prediction against the HELD encoder state on hidden slot-frames -- a diagnostic, not a target"
+  },
+  "vis_mask_actions": {
+    "lo": 0,
+    "hi": 64,
+    "rule": "both",
+    "hidden": "token",
+    "dropout": 0.5,
+    "mode": "sequence"
+  },
+  "selected_on": "mse_return (first visible frame after a hide)"
+}

movi_c/dynamics/dynloc_slotcontrast_step30000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc900d5468ea740f2112febda02abcbc843e5390a9f6ae377039ac913341bdd4
+size 4415797

movi_c/dynamics/ocvp_act_isee_result.json

@@ -0,0 +1,221 @@
+{
+  "arm": "t30k_ocvp_act_adaln_mlp2_nosal",
+  "arm_type": "ocvp_par",
+  "params": 1092226,
+  "seed": 0,
+  "history": 14,
+  "rollout_train": 10,
+  "rollout_eval": 10,
+  "layers": 4,
+  "d_model": 128,
+  "ffn": 256,
+  "heads": 4,
+  "grad_clip": 0.0,
+  "batch": 128,
+  "slot_cache": "/group/awm/slotcontrast/cache/movi_c_fix_ours_nosal_slots_v1",
+  "slot_cache_ckpt_step": 97500,
+  "residual": true,
+  "is_null": false,
+  "lr": 0.0002,
+  "schedule": "cosine",
+  "best_ckpt_step": 17500,
+  "ckpt_mismatch": true,
+  "steps": 30000,
+  "val_mse_per_step": [
+    0.008146516978740692,
+    0.018734635785222054,
+    0.029591821134090424,
+    0.04022786393761635,
+    0.04973062872886658,
+    0.05926008149981499,
+    0.06791070103645325,
+    0.07618439197540283,
+    0.08428974449634552,
+    0.09179138392210007
+  ],
+  "val_mse_mean": 0.05258677527308464,
+  "val_n": 250,
+  "best_val_mse": 0.05254486948251724,
+  "val_mse_per_step_best_ckpt": [
+    0.008163431659340858,
+    0.018768474459648132,
+    0.029623540118336678,
+    0.04023287445306778,
+    0.049720872193574905,
+    0.059219423681497574,
+    0.06783756613731384,
+    0.07608082890510559,
+    0.08416438102722168,
+    0.0916372761130333
+  ],
+  "val_mse_mean_best_ckpt": 0.05254486948251724,
+  "reported_checkpoint": "latest.pt (matches sc_dyn_eval.py)",
+  "history_curve": [
+    {
+      "step": 10000,
+      "val_mse_mean": 0.053050845861434937,
+      "val_sel_mean": 0.053050845861434937,
+      "val_mse_per_step": [
+        0.008255324326455593,
+        0.018988922238349915,
+        0.02998383715748787,
+        0.04072330519556999,
+        0.05030427128076553,
+        0.05984983593225479,
+        0.06853798031806946,
+        0.07682956010103226,
+        0.08484499156475067,
+        0.09219043701887131
+      ],
+      "improved": true
+    },
+    {
+      "step": 12500,
+      "val_mse_mean": 0.052621107548475266,
+      "val_sel_mean": 0.052621107548475266,
+      "val_mse_per_step": [
+        0.00820538867264986,
+        0.018853234127163887,
+        0.029733238741755486,
+        0.04036162793636322,
+        0.049856510013341904,
+        0.059339992702007294,
+        0.06794542819261551,
+        0.07615538686513901,
+        0.08416497707366943,
+        0.09159529954195023
+      ],
+      "improved": true
+    },
+    {
+      "step": 15000,
+      "val_mse_mean": 0.05256000906229019,
+      "val_sel_mean": 0.05256000906229019,
+      "val_mse_per_step": [
+        0.00818153191357851,
+        0.018814299255609512,
+        0.02968592941761017,
+        0.040313348174095154,
+        0.049791526049375534,
+        0.05928114429116249,
+        0.06788333505392075,
+        0.07607464492321014,
+        0.0840858444571495,
+        0.09148846566677094
+      ],
+      "improved": true
+    },
+    {
+      "step": 17500,
+      "val_mse_mean": 0.05254486948251724,
+      "val_sel_mean": 0.05254486948251724,
+      "val_mse_per_step": [
+        0.008163431659340858,
+        0.018768474459648132,
+        0.029623540118336678,
+        0.04023287445306778,
+        0.049720872193574905,
+        0.059219423681497574,
+        0.06783756613731384,
+        0.07608082890510559,
+        0.08416438102722168,
+        0.0916372761130333
+      ],
+      "improved": true
+    },
+    {
+      "step": 20000,
+      "val_mse_mean": 0.05269046500325203,
+      "val_sel_mean": 0.05269046500325203,
+      "val_mse_per_step": [
+        0.008157257921993732,
+        0.018768519163131714,
+        0.02964559942483902,
+        0.04030122607946396,
+        0.04982249066233635,
+        0.05938591808080673,
+        0.06805853545665741,
+        0.07634010165929794,
+        0.08446326106786728,
+        0.09196174889802933
+      ],
+      "improved": false
+    },
+    {
+      "step": 22500,
+      "val_mse_mean": 0.052642665803432465,
+      "val_sel_mean": 0.052642665803432465,
+      "val_mse_per_step": [
+        0.008153584785759449,
+        0.01874658651649952,
+        0.029613306745886803,
+        0.040251195430755615,
+        0.04976711422204971,
+        0.059306152164936066,
+        0.0679701641201973,
+        0.07628575712442398,
+        0.08440346270799637,
+        0.0919293612241745
+      ],
+      "improved": false
+    },
+    {
+      "step": 25000,
+      "val_mse_mean": 0.05265665054321289,
+      "val_sel_mean": 0.05265665054321289,
+      "val_mse_per_step": [
+        0.008152695372700691,
+        0.018750594928860664,
+        0.029617076739668846,
+        0.040262334048748016,
+        0.04978596419095993,
+        0.05934678018093109,
+        0.06800610572099686,
+        0.07629399746656418,
+        0.08441455662250519,
+        0.09193641692399979
+      ],
+      "improved": false
+    },
+    {
+      "step": 27500,
+      "val_mse_mean": 0.05260379984974861,
+      "val_sel_mean": 0.05260379984974861,
+      "val_mse_per_step": [
+        0.008144566789269447,
+        0.018729781731963158,
+        0.02958732284605503,
+        0.04022650048136711,
+        0.04973769187927246,
+        0.059274230152368546,
+        0.0679287239909172,
+        0.07621831446886063,
+        0.08433664590120316,
+        0.09185423702001572
+      ],
+      "improved": false
+    },
+    {
+      "step": 30000,
+      "val_mse_mean": 0.05258677527308464,
+      "val_sel_mean": 0.05258677527308464,
+      "val_mse_per_step": [
+        0.008146516978740692,
+        0.018734635785222054,
+        0.029591821134090424,
+        0.04022786393761635,
+        0.04973062872886658,
+        0.05926008149981499,
+        0.06791070103645325,
+        0.07618439197540283,
+        0.08428974449634552,
+        0.09179138392210007
+      ],
+      "improved": false
+    }
+  ],
+  "stop_reason": "step budget reached",
+  "wallclock_s": 19811.8,
+  "supervise": "dense",
+  "vis_mask": false
+}

movi_c/dynamics/ocvp_act_isee_step30000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecd98730576d499cea8098ae1a39e5a4148e5b6afdb5b84c50109a6e4f1cc16d
+size 4399605

movi_c/dynamics/ocvp_act_slotcontrast_result.json

@@ -0,0 +1,221 @@
+{
+  "arm": "t30k_ocvp_act_adaln_mlp2full_sc",
+  "arm_type": "ocvp_par",
+  "params": 1096066,
+  "seed": 0,
+  "history": 14,
+  "rollout_train": 10,
+  "rollout_eval": 10,
+  "layers": 4,
+  "d_model": 128,
+  "ffn": 256,
+  "heads": 4,
+  "grad_clip": 0.0,
+  "batch": 128,
+  "slot_cache": "/group/awm/slotcontrast/cache/movi_c_sc_slots_v1",
+  "slot_cache_ckpt_step": 75000,
+  "residual": true,
+  "is_null": false,
+  "lr": 0.0002,
+  "schedule": "cosine",
+  "best_ckpt_step": 27500,
+  "ckpt_mismatch": true,
+  "steps": 30000,
+  "val_mse_per_step": [
+    0.003849407657980919,
+    0.00913578923791647,
+    0.01494532823562622,
+    0.020837923511862755,
+    0.026366135105490685,
+    0.03174756467342377,
+    0.03700023144483566,
+    0.04227001592516899,
+    0.04727056249976158,
+    0.05222862586379051
+  ],
+  "val_mse_mean": 0.02856515720486641,
+  "val_n": 250,
+  "best_val_mse": 0.028559476137161255,
+  "val_mse_per_step_best_ckpt": [
+    0.003849732456728816,
+    0.009135734289884567,
+    0.014943580143153667,
+    0.020832745358347893,
+    0.02635871060192585,
+    0.03173966705799103,
+    0.03699290007352829,
+    0.042262040078639984,
+    0.04726221039891243,
+    0.052217431366443634
+  ],
+  "val_mse_mean_best_ckpt": 0.028559476137161255,
+  "reported_checkpoint": "latest.pt (matches sc_dyn_eval.py)",
+  "history_curve": [
+    {
+      "step": 10000,
+      "val_mse_mean": 0.029168466106057167,
+      "val_sel_mean": 0.029168466106057167,
+      "val_mse_per_step": [
+        0.00389186036773026,
+        0.00931267999112606,
+        0.015285424888134003,
+        0.021340064704418182,
+        0.02703428640961647,
+        0.03255582973361015,
+        0.03786364570260048,
+        0.043168168514966965,
+        0.04817553609609604,
+        0.05305715650320053
+      ],
+      "improved": true
+    },
+    {
+      "step": 12500,
+      "val_mse_mean": 0.028875086456537247,
+      "val_sel_mean": 0.028875086456537247,
+      "val_mse_per_step": [
+        0.0038741901516914368,
+        0.009233511984348297,
+        0.015147539786994457,
+        0.021135160699486732,
+        0.026751872152090073,
+        0.032202210277318954,
+        0.03745617717504501,
+        0.04271648824214935,
+        0.04767796769738197,
+        0.05255574733018875
+      ],
+      "improved": true
+    },
+    {
+      "step": 15000,
+      "val_mse_mean": 0.02867993153631687,
+      "val_sel_mean": 0.02867993153631687,
+      "val_mse_per_step": [
+        0.003855296643450856,
+        0.009176499210298061,
+        0.015030371956527233,
+        0.020967749878764153,
+        0.0265472661703825,
+        0.03194752335548401,
+        0.037167370319366455,
+        0.04242616519331932,
+        0.04738939180970192,
+        0.05229167640209198
+      ],
+      "improved": true
+    },
+    {
+      "step": 17500,
+      "val_mse_mean": 0.028664615005254745,
+      "val_sel_mean": 0.028664615005254745,
+      "val_mse_per_step": [
+        0.0038506726268678904,
+        0.009165704250335693,
+        0.015001067891716957,
+        0.02092078886926174,
+        0.026489151641726494,
+        0.03189229965209961,
+        0.03714843466877937,
+        0.04242745414376259,
+        0.04741044342517853,
+        0.052340149879455566
+      ],
+      "improved": true
+    },
+    {
+      "step": 20000,
+      "val_mse_mean": 0.028592247515916824,
+      "val_sel_mean": 0.028592247515916824,
+      "val_mse_per_step": [
+        0.0038518342189490795,
+        0.009153453633189201,
+        0.0149763198569417,
+        0.020885378122329712,
+        0.026422550901770592,
+        0.03180137649178505,
+        0.0370357371866703,
+        0.042297083884477615,
+        0.04727582633495331,
+        0.05222291871905327
+      ],
+      "improved": true
+    },
+    {
+      "step": 22500,
+      "val_mse_mean": 0.028617363423109055,
+      "val_sel_mean": 0.028617363423109055,
+      "val_mse_per_step": [
+        0.003850064007565379,
+        0.009149616584181786,
+        0.014965064823627472,
+        0.020871305838227272,
+        0.026417246088385582,
+        0.03181121498346329,
+        0.037071507424116135,
+        0.042352184653282166,
+        0.04736169055104256,
+        0.05232373625040054
+      ],
+      "improved": false
+    },
+    {
+      "step": 25000,
+      "val_mse_mean": 0.028581123799085617,
+      "val_sel_mean": 0.028581123799085617,
+      "val_mse_per_step": [
+        0.0038489135913550854,
+        0.009140389040112495,
+        0.014953809790313244,
+        0.02085118740797043,
+        0.026386022567749023,
+        0.03176923096179962,
+        0.037024155259132385,
+        0.04229341819882393,
+        0.04729386419057846,
+        0.052250251173973083
+      ],
+      "improved": true
+    },
+    {
+      "step": 27500,
+      "val_mse_mean": 0.028559476137161255,
+      "val_sel_mean": 0.028559476137161255,
+      "val_mse_per_step": [
+        0.003849732456728816,
+        0.009135734289884567,
+        0.014943580143153667,
+        0.020832745358347893,
+        0.02635871060192585,
+        0.03173966705799103,
+        0.03699290007352829,
+        0.042262040078639984,
+        0.04726221039891243,
+        0.052217431366443634
+      ],
+      "improved": true
+    },
+    {
+      "step": 30000,
+      "val_mse_mean": 0.02856515720486641,
+      "val_sel_mean": 0.02856515720486641,
+      "val_mse_per_step": [
+        0.003849407657980919,
+        0.00913578923791647,
+        0.01494532823562622,
+        0.020837923511862755,
+        0.026366135105490685,
+        0.03174756467342377,
+        0.03700023144483566,
+        0.04227001592516899,
+        0.04727056249976158,
+        0.05222862586379051
+      ],
+      "improved": false
+    }
+  ],
+  "stop_reason": "step budget reached",
+  "wallclock_s": 19767.2,
+  "supervise": "dense",
+  "vis_mask": false
+}

movi_c/dynamics/ocvp_act_slotcontrast_step30000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25ae820604024ef7583dab1c2e8dce9492e0b0cf76a0990cf69372e76ae2b9b4
+size 4414965

movi_c/isee/encoder_best.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e77a0d3c5d782a01047932a869cf891ee0c83086dde066a9653505fae0ca744e
+size 36536609

movi_c/isee/encoder_best.live.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98db5e0aba6d0722837a8d4b6e676203fb1d88dcddd3233b6a1e5af392a82bf6
+size 98459531

movi_c/isee/settings.yaml

@@ -0,0 +1,235 @@
+optimizer:
+  name: Adam
+  lr: 0.0004
+  lr_scheduler:
+    name: exp_decay_with_warmup
+    warmup_steps: 2500
+    decay_steps: 100000
+model:
+  initializer:
+    name: FixedLearnedInit
+    n_slots: 11
+    dim: 64
+  encoder:
+    backbone:
+      name: CachedFeatures
+      dim: 384
+      n_patches: 576
+    output_transform:
+      name: networks.two_layer_mlp
+      inp_dim: 384
+      outp_dim: 64
+      hidden_dim: 768
+      layer_norm: true
+    post_pos_embed:
+      name: ConcatPatchPosEmbed
+      n_patches: 576
+      mode: fourier
+      bands: 1
+      include_raw: false
+      norm_features: false
+  grouper:
+    name: SlotAttention
+    inp_dim: 68
+    slot_dim: 64
+    n_iters: 2
+    use_mlp: true
+    stream_split: 60
+    inp_split: 64
+    pos_logit_gain: 1.0
+    pos_norm: rms
+    pos_mlp: false
+    pos_init: identity
+  decoder:
+    name: CanonMLPDecoder
+    inp_dim: 64
+    outp_dim: 384
+    hidden_dims:
+    - 1024
+    - 1024
+    - 1024
+    n_patches: 576
+    pose_dims:
+    - 60
+    - 64
+    content_dims:
+    - 0
+    - 60
+    pose_hidden: 64
+    scale_range:
+    - 0.01
+    - 4.0
+    abs_table: false
+    h_init: default
+    h_hidden_dims: []
+    fourier_bands: 8
+    delta: 5.0
+    pos_gain: 1.0
+    gate_mode: none
+  dynamics_predictor: null
+  predictor:
+    name: networks.SplitSlotPredictor
+    dim: 64
+    content_dims:
+    - 0
+    - 60
+    pose_dims:
+    - 60
+    - 64
+    appearance_blocks: 1
+    appearance_heads: 4
+    kinematics_blocks: 1
+    kinematics_heads: 4
+    content_cond_dim: 8
+    detach_content_cond: true
+  target_encoder: null
+  latent_processor:
+    first_step_corrector_args:
+      n_iters: 3
+  mask_resizers: null
+  losses:
+    loss_featrec:
+      name: MSELoss
+      pred_dims:
+      - 0
+      - 384
+    loss_ss:
+      name: Slot_Slot_Contrastive_Loss
+      pred_key: processor.state
+      temperature: 0.1
+      batch_contrast: true
+      patch_inputs: false
+      keep_input_dim: true
+      dims:
+      - 0
+      - 60
+  loss_weights:
+    loss_featrec: 1.0
+    loss_ss: 0.5
+  input_type: video
+  target_type: features
+  target_encoder_input: null
+  encoder_input: features
+  decoder_source: state
+  aux_decoders: null
+  pointmap: null
+  cross_view: null
+  occlusion: null
+  anchor_supervision: null
+  visualize: false
+  eval_mode_config: null
+  visualize_every_n_steps: 10000
+  masks_to_visualize: null
+  load_weights: null
+  modules_to_load: null
+dataset:
+  train_shards: movi_c/movi_c-train-{000000..000304}.tar
+  val_shards: movi_c/movi_c-validation-{000000..000001}.tar
+  batch_size: 64
+  val_batch_size: 4
+  val_size: 64
+  num_workers: 16
+  num_val_workers: 2
+  train_pipeline:
+    video_size: 24
+    chunk_size: 4
+    sample_one_chunk_per_video: true
+    keys:
+    - video
+    shuffle_size: 512
+    transforms:
+      name: movi_train
+      type: video
+      input_size: 336
+      h_flip_prob: 0.5
+    name: CachedViewPipeline
+    cache:
+      path: /group/awm/datasets/viewcache/movi_c/train_nope_s14_336
+      standardize: false
+      n_patches: 576
+      dim: 384
+  val_pipeline:
+    use_chunks: false
+    keys:
+    - video
+    - segmentations
+    transforms:
+      name: movi_val
+      type: video
+      input_size: 336
+      num_classes: 11
+    name: CachedViewPipeline
+    cache:
+      path: /group/awm/datasets/viewcache/movi_c/val_nope_s14_336
+      standardize: false
+      n_patches: 576
+      dim: 384
+trainer:
+  max_steps: 100000
+  log_every_n_steps: 250
+  val_check_interval: 2500
+  gradient_clip_val: 0.05
+  accelerator: auto
+train_metrics: null
+val_metrics:
+  ari:
+    name: VideoARI
+    ignore_background: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  image_ari:
+    name: ImageARI
+    ignore_background: true
+    video_input: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  mbo:
+    name: VideoIoU
+    matching: overlap
+    ignore_background: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  image_mbo:
+    name: ImageIoU
+    matching: overlap
+    ignore_background: true
+    video_input: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+activation_reg_schedule: null
+depth_beta_schedule: null
+globals:
+  NUM_SLOTS: 11
+  SLOT_DIM: 64
+  DINO_MODEL: vit_small_patch14_dinov2
+  FEAT_DIM: 384
+  NUM_PATCHES: 576
+  NUM_GPUS: 1
+  BATCH_SIZE_PER_GPU: 64
+  TOTAL_BATCH_SIZE: 64
+  BASE_LR: 0.0001
+  POS_DIM: 4
+  BANDS: 1
+  CONTENT_LO: 0
+  CONTENT_HI: 60
+  POSE_LO: 60
+  POSE_HI: 64
+experiment_name: paper_movi_c_fix_ours_nosal_vc
+experiment_group: slotcontrast
+seed: 20260826
+checkpoint_every_n_steps: 2500
+checkpoint_save_top_k: -1
+checkpoint_best:
+  monitor: val/sel
+  mode: max
+wandb:
+  project: tsocr_table1
+  entity: pramish-paudel-insait
+  group: movi_c
+  tags:
+  - ours_nosal
+  - movi_c
+  - fix
+  - vits
+  - stage1
+  - vc

movi_c/slotcontrast/best.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bfabb88d239e22b6230b41a9543079be345ca6725bc0a163674e04a677a75e7
+size 36832851

movi_c/slotcontrast/best.live.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fad1603f3a4ddc509eb9fcea1176cfaeefb4c39fab7c9e6000e19c3013a1d72
+size 100583272

movi_c/slotcontrast/settings.yaml

@@ -0,0 +1,184 @@
+optimizer:
+  name: Adam
+  lr: 0.0004
+  lr_scheduler:
+    name: exp_decay_with_warmup
+    warmup_steps: 2500
+    decay_steps: 100000
+model:
+  initializer:
+    name: FixedLearnedInit
+    n_slots: 11
+    dim: 64
+  encoder:
+    backbone:
+      name: CachedFeatures
+      dim: 384
+      n_patches: 576
+    output_transform:
+      name: networks.two_layer_mlp
+      inp_dim: 384
+      outp_dim: 64
+      hidden_dim: 768
+      layer_norm: true
+  grouper:
+    name: SlotAttention
+    inp_dim: 64
+    slot_dim: 64
+    n_iters: 2
+    use_mlp: true
+  decoder:
+    name: MLPDecoder
+    inp_dim: 64
+    outp_dim: 384
+    hidden_dims:
+    - 1024
+    - 1024
+    - 1024
+    n_patches: 576
+  dynamics_predictor: null
+  predictor:
+    name: networks.TransformerEncoder
+    dim: 64
+    n_blocks: 1
+    n_heads: 4
+  target_encoder: null
+  latent_processor:
+    first_step_corrector_args:
+      n_iters: 3
+  mask_resizers: null
+  losses:
+    loss_featrec:
+      name: MSELoss
+      pred_dims:
+      - 0
+      - 384
+    loss_ss:
+      name: Slot_Slot_Contrastive_Loss
+      pred_key: processor.state
+      temperature: 0.1
+      batch_contrast: true
+      patch_inputs: false
+      keep_input_dim: true
+  loss_weights:
+    loss_featrec: 1.0
+    loss_ss: 0.5
+  input_type: video
+  target_type: features
+  target_encoder_input: null
+  encoder_input: features
+  decoder_source: state
+  aux_decoders: null
+  pointmap: null
+  cross_view: null
+  occlusion: null
+  anchor_supervision: null
+  visualize: false
+  eval_mode_config: null
+  visualize_every_n_steps: 10000
+  masks_to_visualize: null
+  load_weights: null
+  modules_to_load: null
+dataset:
+  train_shards: movi_c/movi_c-train-{000000..000304}.tar
+  val_shards: movi_c/movi_c-validation-{000000..000001}.tar
+  batch_size: 64
+  val_batch_size: 4
+  val_size: 64
+  num_workers: 16
+  num_val_workers: 2
+  train_pipeline:
+    video_size: 24
+    chunk_size: 4
+    sample_one_chunk_per_video: true
+    keys:
+    - video
+    shuffle_size: 512
+    transforms:
+      name: movi_train
+      type: video
+      input_size: 336
+      h_flip_prob: 0.5
+    name: CachedViewPipeline
+    cache:
+      path: /group/awm/datasets/viewcache/movi_c/train_dino_s14_336
+      standardize: false
+      n_patches: 576
+      dim: 384
+  val_pipeline:
+    use_chunks: false
+    keys:
+    - video
+    - segmentations
+    transforms:
+      name: movi_val
+      type: video
+      input_size: 336
+      num_classes: 11
+    name: CachedViewPipeline
+    cache:
+      path: /group/awm/datasets/viewcache/movi_c/val_dino_s14_336
+      standardize: false
+      n_patches: 576
+      dim: 384
+trainer:
+  max_steps: 100000
+  log_every_n_steps: 250
+  val_check_interval: 2500
+  gradient_clip_val: 0.05
+  accelerator: auto
+train_metrics: null
+val_metrics:
+  ari:
+    name: VideoARI
+    ignore_background: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  image_ari:
+    name: ImageARI
+    ignore_background: true
+    video_input: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  mbo:
+    name: VideoIoU
+    matching: overlap
+    ignore_background: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+  image_mbo:
+    name: ImageIoU
+    matching: overlap
+    ignore_background: true
+    video_input: true
+    pred_key: decoder_masks_hard
+    true_key: segmentations
+activation_reg_schedule: null
+depth_beta_schedule: null
+globals:
+  NUM_SLOTS: 11
+  SLOT_DIM: 64
+  DINO_MODEL: vit_small_patch14_dinov2
+  FEAT_DIM: 384
+  NUM_PATCHES: 576
+  NUM_GPUS: 1
+  BATCH_SIZE_PER_GPU: 64
+  TOTAL_BATCH_SIZE: 64
+  BASE_LR: 0.0001
+experiment_name: paper_movi_c_sc_vc
+experiment_group: slotcontrast
+seed: 20260826
+checkpoint_every_n_steps: 2500
+checkpoint_save_top_k: -1
+checkpoint_best:
+  monitor: val/sel
+  mode: max
+wandb:
+  project: tsocr_table1
+  entity: pramish-paudel-insait
+  group: movi_c
+  tags:
+  - sc
+  - movi_c
+  - stage1
+  - vc