Add selected weights for D4P_PANDA (3 files)

selected/D4P_PANDA/config/01__config.json

@@ -0,0 +1,165 @@
+{
+  "dataset": "K360",
+  "base_path": "<REMOTE_PATH>",
+  "batch_size": 32,
+  "epochs": 35,
+  "learning_rate": 0.0003,
+  "lr_scheduler": "none",
+  "lr_step": 7,
+  "lr_gamma": 0.4,
+  "seed": 0,
+  "shuffle": false,
+  "num_workers": 0,
+  "folder_name": "pandaset_bb0_fine_seed0_corev2b",
+  "checkpoint_root": "<REMOTE_PATH>",
+  "log_dir": null,
+  "resume": null,
+  "limit_batches": null,
+  "limit_val": null,
+  "device": "cuda",
+  "use_features": [
+    "class",
+    "color",
+    "position",
+    "num"
+  ],
+  "class_embed": false,
+  "color_embed": false,
+  "pointnet_path": "<REMOTE_PATH>",
+  "pointnet_layers": 3,
+  "pointnet_variation": 0,
+  "pointnet_numpoints": 256,
+  "pointnet_freeze": false,
+  "pointnet_features": 2,
+  "num_stats_path": "<REMOTE_PATH>",
+  "no_pc_augment": true,
+  "object_size": 28,
+  "pad_size": 16,
+  "num_mentioned": 6,
+  "coarse_embed_dim": 256,
+  "mfam_heads": 4,
+  "bilstm_stride": 2,
+  "bilstm_layers": 1,
+  "ste_word_layers": 1,
+  "ste_inter_layers": 1,
+  "ste_heads": 4,
+  "prior_layers": 2,
+  "prior_heads": 8,
+  "dropout": 0.1,
+  "fine_embed_dim": 128,
+  "fine_text_layers": 1,
+  "cra_heads": 4,
+  "cra_steps": 6,
+  "pmc_prob": 0.5,
+  "pmc_threshold": 0.4,
+  "offset_lambda": 5.0,
+  "warmup_epochs": 3,
+  "regressor_cell": "all",
+  "regressor_learn": "center",
+  "regressor_eval": "center",
+  "t5_path": "<REMOTE_PATH>",
+  "clip_path": null,
+  "ste_stage1_path": "<REMOTE_PATH>",
+  "clip_dim": 512,
+  "max_text_len": 32,
+  "hungging_model": null,
+  "fixed_embedding": false,
+  "ranking_loss": "contrastive",
+  "temperature": 0.1,
+  "top_k": [
+    1,
+    3,
+    5,
+    10,
+    25
+  ],
+  "select_k": 5,
+  "allow_test": false,
+  "implementation": "Des4Pos paper reimplementation (not authors' code)",
+  "known_classes": [
+    "animal",
+    "barrier",
+    "bicycle",
+    "bus",
+    "car",
+    "construction",
+    "motorcycle",
+    "pedestrian",
+    "pushable pullable",
+    "sign",
+    "trafficcone",
+    "trailer",
+    "train",
+    "truck",
+    "vehicle"
+  ],
+  "params_total": 13408192,
+  "params_trainable": 2374562,
+  "source_sha256": {},
+  "n_train_queries": 4000,
+  "n_val_queries": 2000,
+  "num_stats": {
+    "core": "<REMOTE_PATH>",
+    "core_manifest_sha256": "c51bf63a535d78f299d6a556560eafb0660f39c8c8dd017efbd680cccd2890fc",
+    "counts_sha256": "b2340c73453587eee80695b0d09d37096c4e36cf48e13b2abd6b67e9285a0ceb",
+    "dataset_manifest_sha256": "04fbea46bbbc82a035378fbb891944361641f03a355fed4954fb7fa704e7a494",
+    "dedup_by_cell_id": true,
+    "domain": "pandaset",
+    "fitted_before_fixedpoints": true,
+    "max_count": 100,
+    "mean": 100.0,
+    "min_count": 100,
+    "n_cells_seen": 380,
+    "n_cells_unique": 380,
+    "n_excluded": 0,
+    "n_objects": 5753,
+    "n_objects_seen": 5753,
+    "n_objects_unique": 5753,
+    "n_seen": 5753,
+    "pinned_zero_variance": true,
+    "provenance": "train_fit",
+    "scene_list_sha256": "651a0fe0d037702d8c7d6fcfbf62d5dadf5699bdd7153a9114dd50c7ae51a848",
+    "source": "<REMOTE_PATH> TRAIN cells (50 scenes, dedup by cell.id); raw obj.xyz before FixedPoints",
+    "split": "train",
+    "std": 1.0,
+    "std_ddof": 0,
+    "version": 1
+  },
+  "num_stats_fingerprint": "1b0a037b53f3aaaa2d410980f0095816de74aed346a132cd368ddeba26ff61bb",
+  "num_stats_source": "<REMOTE_PATH>",
+  "object_encoder_factory": {
+    "pointnet_head_classes": 22,
+    "pointnet_head_colors": 8,
+    "target_num_classes": 15,
+    "target_num_colors": 8,
+    "num_stats": {
+      "core": "<REMOTE_PATH>",
+      "core_manifest_sha256": "c51bf63a535d78f299d6a556560eafb0660f39c8c8dd017efbd680cccd2890fc",
+      "counts_sha256": "b2340c73453587eee80695b0d09d37096c4e36cf48e13b2abd6b67e9285a0ceb",
+      "dataset_manifest_sha256": "04fbea46bbbc82a035378fbb891944361641f03a355fed4954fb7fa704e7a494",
+      "dedup_by_cell_id": true,
+      "domain": "pandaset",
+      "fitted_before_fixedpoints": true,
+      "max_count": 100,
+      "mean": 100.0,
+      "min_count": 100,
+      "n_cells_seen": 380,
+      "n_cells_unique": 380,
+      "n_excluded": 0,
+      "n_objects": 5753,
+      "n_objects_seen": 5753,
+      "n_objects_unique": 5753,
+      "n_seen": 5753,
+      "pinned_zero_variance": true,
+      "provenance": "train_fit",
+      "scene_list_sha256": "651a0fe0d037702d8c7d6fcfbf62d5dadf5699bdd7153a9114dd50c7ae51a848",
+      "source": "<REMOTE_PATH> TRAIN cells (50 scenes, dedup by cell.id); raw obj.xyz before FixedPoints",
+      "split": "train",
+      "std": 1.0,
+      "std_ddof": 0,
+      "version": 1
+    },
+    "num_stats_fingerprint": "1b0a037b53f3aaaa2d410980f0095816de74aed346a132cd368ddeba26ff61bb",
+    "classifier_outputs_consumed": false
+  }
+}

selected/D4P_PANDA/normalization/NORMALIZATION.json

@@ -0,0 +1,107 @@
+{
+  "mean": [
+    0.5352351069450378,
+    0.5,
+    0.5463622212409973,
+    0.5,
+    0.5,
+    0.1448325216770172,
+    0.1448955237865448,
+    0.23320449888706207,
+    0.5000000596046448,
+    0.0017008326249197125,
+    -0.009016148746013641,
+    0.08206436783075333,
+    0.23505154252052307,
+    0.18795736134052277,
+    0.5,
+    0.5000032782554626,
+    0.5,
+    0.38766154646873474,
+    0.5458333492279053,
+    0.47092947363853455,
+    0.40427646040916443,
+    1.5081766843795776,
+    1.4996931552886963
+  ],
+  "std": [
+    0.12605366110801697,
+    0.30047550797462463,
+    0.2683347165584564,
+    0.30047550797462463,
+    0.30047550797462463,
+    0.35188278555870056,
+    0.12740030884742737,
+    0.34719395637512207,
+    0.30047550797462463,
+    0.07191990315914154,
+    0.06962312012910843,
+    0.058048903942108154,
+    0.3480305075645447,
+    0.2830078899860382,
+    0.30047550797462463,
+    0.2901189923286438,
+    0.30047550797462463,
+    0.1723688840866089,
+    0.26841557025909424,
+    0.6198450922966003,
+    0.6384478807449341,
+    0.9944097399711609,
+    0.9927133321762085
+  ],
+  "pinned": [
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false,
+    false
+  ],
+  "pinned_names": [],
+  "feature_names": [
+    "cos",
+    "cos_rank",
+    "sp",
+    "p0_rank",
+    "p2_rank",
+    "is_tied",
+    "tied_size",
+    "sc_center",
+    "sc_center_rank",
+    "off_dx",
+    "off_dy",
+    "off_norm",
+    "sc_fine",
+    "sat_fine",
+    "sc_fine_rank",
+    "borda",
+    "c_rank",
+    "n_obj",
+    "n_matched",
+    "dist_nearest_matched",
+    "dist_centroid_matched",
+    "pred_dist_anchor",
+    "center_dist_anchor"
+  ],
+  "feature_version": "FEATURE_COORD_FIX_v1",
+  "fitted_on": "TRAIN valid lists only, flattened 23 features, std<1e-3 pinned to 1",
+  "train_valid_lists": 3075,
+  "train_npz_sha256": "3e9caf72b96f710306f55494f76ad3ca43907cce784410d45c6de8733f74e4f8"
+}

selected/D4P_PANDA/scar_head/ht23_pandaset_s0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdc0822ba3516506fcd5193afb7e45fda894fcaa3f836e0083b29921c903ea13
+size 28898