pimm export (config.json)

README.md

@@ -1,27 +1,16 @@
 ---
-license: mit
 library_name: pimm
-datasets:
-- DeepLearnPhysics/PILArNet-M
 tags:
-- particle
-- physics
-- 3D
-- lartpc
-- panda
-- point-transformer-v3
+  - particle-physics
+  - lartpc
+  - point-cloud
 ---
 
-
 # Panda — interaction panoptic detector
 
-A query-based panoptic detector that groups LArTPC points into
-interactions (Mask2Former / OneFormer3D-style masked cross-attention decoder over a
-PTv3 backbone), with a per-query classification head and an auxiliary stuff head.
+Query-based panoptic detector grouping LArTPC points into interactions (masked cross-attention decoder over a PTv3 backbone).
 
-- **pimm type:** `detector-v4`
-- **queries:** 12 · **classes:** 2 · **stuff classes:** [0]
-- **decoder depth:** 3 · **hidden:** 256 · **heads:** 16
+- **pimm type:** `detector-v4` · 12 queries · 2 classes
 
 ## Loading
 
@@ -30,24 +19,9 @@ import pimm
 model = pimm.from_pretrained("hf://deeplearnphysics/panda-interaction")
 ```
 
-The architecture and all hyper-parameters are carried in `training_config.json`,
-so no config file is needed. Weights are bitwise-identical to the original
-checkpoint.
-
-## Input
-
-A pimm `Point` dict for LArTPC (PILArNet) point clouds: `coord` (xyz) + `energy`,
-collected as `feat_keys=("coord", "energy")` (4 input channels). Coordinates are
-normalized in the model forward.
-
-## Notes
-
-- Config sets `enable_flash=True`. On CPU or non-FlashAttention setups, override
-  it (e.g. pass `enable_flash=False` to `from_pretrained`, or for detectors set it
-  on the `backbone`).
+Architecture + hyper-parameters travel in `config.json`, so no config file is
+needed. Weights are bitwise-identical to the original checkpoint.
 
 ## Provenance
 
-Repackaged from the original [`DeepLearnPhysics/panda`](https://github.com/DeepLearnPhysics/panda) checkpoints into the
-[`pimm`](https://github.com/youngsm/particle-imaging-models) export format. Inherits
-the license of the source repository.
+Repackaged from the original [`panda`](https://github.com/DeepLearnPhysics/panda) checkpoints into the [`pimm`](https://github.com/youngsm/particle-imaging-models) export format. Inherits the source repo's license.

config.json

@@ -0,0 +1,109 @@
+{
+  "model": {
+    "type": "detector-v4",
+    "labels": [
+      "interaction"
+    ],
+    "num_queries": 12,
+    "num_classes": 2,
+    "use_stuff_head": true,
+    "stuff_classes": [
+      0
+    ],
+    "mlp_point_proj": false,
+    "full_in_channels": 1232,
+    "hidden_channels": 256,
+    "num_heads": 16,
+    "depth": 3,
+    "mlp_ratio": 4.0,
+    "qkv_bias": true,
+    "qk_scale": null,
+    "layer_scale": null,
+    "pre_norm": true,
+    "query_type": "learned",
+    "pos_emb": true,
+    "supervise_attn_mask": true,
+    "enable_flash": false,
+    "upcast_attention": false,
+    "upcast_softmax": false,
+    "backbone": {
+      "type": "PT-v3m2",
+      "in_channels": 4,
+      "order": [
+        "hilbert",
+        "hilbert-trans",
+        "z",
+        "z-trans"
+      ],
+      "stride": [
+        2,
+        2,
+        2,
+        2
+      ],
+      "enc_depths": [
+        3,
+        3,
+        3,
+        9,
+        3
+      ],
+      "enc_channels": [
+        48,
+        96,
+        192,
+        384,
+        512
+      ],
+      "enc_num_head": [
+        3,
+        6,
+        12,
+        24,
+        32
+      ],
+      "enc_patch_size": [
+        256,
+        256,
+        256,
+        256,
+        256
+      ],
+      "mlp_ratio": 4,
+      "qkv_bias": true,
+      "qk_scale": null,
+      "layer_scale": 1e-05,
+      "attn_drop": 0.0,
+      "proj_drop": 0.0,
+      "drop_path": 0.3,
+      "shuffle_orders": true,
+      "pre_norm": true,
+      "enable_rpe": false,
+      "enable_flash": true,
+      "upcast_attention": false,
+      "upcast_softmax": false,
+      "traceable": false,
+      "mask_token": false,
+      "enc_mode": true,
+      "freeze_encoder": false
+    },
+    "criteria": [
+      {
+        "type": "FastUnifiedInstanceLoss",
+        "cost_mask": 1.0,
+        "cost_dice": 1.0,
+        "cost_class": 1.0,
+        "loss_weight_focal": 2.0,
+        "loss_weight_dice": 5.0,
+        "cls_weight_matched": 2.0,
+        "cls_weight_noobj": 0.5,
+        "focal_alpha": 0.25,
+        "focal_gamma": 2.0,
+        "aux_loss_weight": 1.0,
+        "num_points": 100000,
+        "truth_label": "instance",
+        "noobj_mask_loss_weight": 0.0
+      }
+    ]
+  }
+}