Fix/fp centroid distance inf (problems with cari4d for video 2, 4, 7)

reconstruction/README.md

@@ -2,6 +2,12 @@
 
 Docker-based modules for video reconstruction, depth estimation, object detection, segmentation, mesh generation, pose tracking, and human body modeling.
 
+For the Track 1 challenge, start with the [input inventory](docs/track1_preflight.md)
+to check episode metadata and prepared input paths before running reconstruction.
+Use the [input preparation runner](docs/track1_prepare.md) to build verified masks
+and static meshes, then the [batch runner](docs/track1_batch.md) to plan and execute
+reconstruction. Both default to planning without starting models.
+
 ## Quickstart (MOGE)
 
 Minimal example using MOGE. From `reconstruction/`:

reconstruction/configs/track1_inputs/README.md

@@ -0,0 +1,34 @@
+# Initial prompts for the Track 1 baseline
+
+`preparation.json` covers the 30 released Track 1 episodes. Each episode uses
+frame zero as the reference, with human ID 0 and object ID 1. The boxes use
+pixel coordinates in the original 1536 × 1152 image. They were selected by
+inspecting the real reference images and magnified object crops, including the
+main interacting person where background people are present.
+
+`source_review.json` binds these annotations to the reviewed source video and
+prompt contents by SHA-256. The source media are not included here. Verify the
+dataset against this record before using the prompts with a different copy.
+These are initial segmentation prompts, not ground-truth masks or accepted
+reconstructions. Inspect the generated overlay images and static meshes before
+the full baseline, especially thin hoops, pan/roller handles and occluded cart
+legs. A geometrically valid box does not prove that SAM2 followed the right
+object throughout the video.
+
+Pass this manifest to the [preparation runner](../../docs/track1_prepare.md):
+
+```bash
+python -m v2d.pipelines.track1_prepare \
+  --preparation_manifest reconstruction/configs/track1_inputs/preparation.json \
+  --dataset_root /vol/data/track_1 \
+  --output_dir /vol/outputs/track1/prepared-001 \
+  --sam2_weights /vol/weights/sam2 \
+  --moge_weights "$MOGE_CHECKPOINT" \
+  --sam3d_weights /vol/weights/sam3d \
+  --image_build_commit "$IMAGE_BUILD_COMMIT" --image_digest "$IMAGE_DIGEST"
+```
+
+This command only creates a plan. Resolve weight paths and image provenance from
+the actual deployment. Adding `--execute --episodes 16` prepares one episode;
+full-roster input preparation requires `--execute --all_episodes`. Neither
+command starts temporal CARI4D reconstruction or submits results to Kaggle.

reconstruction/configs/track1_inputs/preparation.json

@@ -0,0 +1,215 @@
+{
+  "schema": "v2d.track1.preparation.v1",
+  "episodes": [
+    {
+      "episode_index": 0,
+      "prompts_path": "prompts/episode_000000.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 1,
+      "prompts_path": "prompts/episode_000001.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 2,
+      "prompts_path": "prompts/episode_000002.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 3,
+      "prompts_path": "prompts/episode_000003.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 4,
+      "prompts_path": "prompts/episode_000004.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 5,
+      "prompts_path": "prompts/episode_000005.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 6,
+      "prompts_path": "prompts/episode_000006.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 7,
+      "prompts_path": "prompts/episode_000007.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 8,
+      "prompts_path": "prompts/episode_000008.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 9,
+      "prompts_path": "prompts/episode_000009.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 10,
+      "prompts_path": "prompts/episode_000010.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 11,
+      "prompts_path": "prompts/episode_000011.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 12,
+      "prompts_path": "prompts/episode_000012.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 13,
+      "prompts_path": "prompts/episode_000013.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 14,
+      "prompts_path": "prompts/episode_000014.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 15,
+      "prompts_path": "prompts/episode_000015.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 16,
+      "prompts_path": "prompts/episode_000016.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 17,
+      "prompts_path": "prompts/episode_000017.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 18,
+      "prompts_path": "prompts/episode_000018.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 19,
+      "prompts_path": "prompts/episode_000019.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 20,
+      "prompts_path": "prompts/episode_000020.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 21,
+      "prompts_path": "prompts/episode_000021.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 22,
+      "prompts_path": "prompts/episode_000022.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 23,
+      "prompts_path": "prompts/episode_000023.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 24,
+      "prompts_path": "prompts/episode_000024.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 25,
+      "prompts_path": "prompts/episode_000025.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 26,
+      "prompts_path": "prompts/episode_000026.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 27,
+      "prompts_path": "prompts/episode_000027.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 28,
+      "prompts_path": "prompts/episode_000028.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    },
+    {
+      "episode_index": 29,
+      "prompts_path": "prompts/episode_000029.json",
+      "reference_frame": 0,
+      "human_id": 0,
+      "object_id": 1
+    }
+  ]
+}

reconstruction/configs/track1_inputs/prompts/episode_000000.json

@@ -0,0 +1,26 @@
+{
+  "prompts": [
+    {
+      "frame_index": 0,
+      "object_id": 0,
+      "role": "human",
+      "box": {
+        "x0": 454,
+        "y0": 384,
+        "x1": 634,
+        "y1": 944
+      }
+    },
+    {
+      "frame_index": 0,
+      "object_id": 1,
+      "role": "object",
+      "box": {
+        "x0": 624,
+        "y0": 771,
+        "x1": 819,
+        "y1": 848
+      }
+    }
+  ]
+}

reconstruction/configs/track1_inputs/prompts/episode_000001.json

@@ -0,0 +1,26 @@
+{
+  "prompts": [
+    {
+      "frame_index": 0,
+      "object_id": 0,
+      "role": "human",
+      "box": {
+        "x0": 771,
+        "y0": 403,
+        "x1": 944,
+        "y1": 896
+      }
+    },
+    {
+      "frame_index": 0,
+      "object_id": 1,
+      "role": "object",
+      "box": {
+        "x0": 586,
+        "y0": 848,
+        "x1": 826,
+        "y1": 982
+      }
+    }
+  ]
+}

reconstruction/configs/track1_inputs/prompts/episode_000002.json

@@ -0,0 +1,26 @@
+{
+  "prompts": [
+    {
+      "frame_index": 0,
+      "object_id": 0,
+      "role": "human",
+      "box": {
+        "x0": 685,
+        "y0": 451,
+        "x1": 810,
+        "y1": 803
+      }
+    },
+    {
+      "frame_index": 0,
+      "object_id": 1,
+      "role": "object",
+      "box": {
+        "x0": 595,
+        "y0": 861,
+        "x1": 861,
+        "y1": 995
+      }
+    }
+  ]
+}

reconstruction/configs/track1_inputs/prompts/episode_000003.json

@@ -0,0 +1,26 @@
+{
+  "prompts": [
+    {
+      "frame_index": 0,
+      "object_id": 0,
+      "role": "human",
+      "box": {
+        "x0": 627,
+        "y0": 406,
+        "x1": 838,
+        "y1": 896
+      }
+    },
+    {
+      "frame_index": 0,
+      "object_id": 1,
+      "role": "object",
+      "box": {
+        "x0": 762,
+        "y0": 643,
+        "x1": 870,
+        "y1": 714
+      }
+    }
+  ]
+}

reconstruction/configs/track1_inputs/prompts/episode_000004.json

@@ -0,0 +1,26 @@
+{
+  "prompts": [
+    {
+      "frame_index": 0,
+      "object_id": 0,
+      "role": "human",
+      "box": {
+        "x0": 803,
+        "y0": 365,
+        "x1": 992,
+        "y1": 902
+      }
+    },
+    {
+      "frame_index": 0,
+      "object_id": 1,
+      "role": "object",
+      "box": {
+        "x0": 650,
+        "y0": 598,
+        "x1": 758,
+        "y1": 666
+      }
+    }
+  ]
+}