Instructions to use EndeavourDD/gnn_wm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use EndeavourDD/gnn_wm with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("EndeavourDD/gnn_wm", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
File size: 13,535 Bytes
4ee0c8c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 | """Self-contained PyG loader for the GNN Disassembly dataset.
Two loader variants:
- load_pyg_frame_products_only(ep, frame) β constraint graph only, no robot
- load_pyg_frame_with_robot(ep, frame) β constraint graph + robot agent node
Both return torch_geometric.data.Data with:
x (N, 268) node features
edge_index (2, N*(N-1)) fully connected directed message-passing edges
edge_attr (N*(N-1), 3) [has_constraint, is_locked, src_blocks_dst]
num_nodes N
Notes on the edge feature design:
- The graph is FULLY CONNECTED and structurally symmetric.
Both (i, j) and (j, i) exist in edge_index for every node pair i != j.
- Direction is NOT encoded in the graph structure. It is encoded as
a feature: `src_blocks_dst`.
- `has_constraint` and `is_locked` are symmetric per pair (same value
for both (i, j) and (j, i)).
- `src_blocks_dst` is asymmetric: it is 1 if the edge's src node
physically blocks its dst node, 0 otherwise.
"""
import json
from dataclasses import dataclass
from pathlib import Path
from typing import Dict, List, Optional, Tuple
import numpy as np
import torch
from torch_geometric.data import Data
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Helpers
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def list_labeled_frames(episode_dir: Path) -> List[int]:
"""Return sorted list of frame indices that have saved annotations."""
mask_dir = episode_dir / "annotations" / "side_masks"
if not mask_dir.exists():
return []
frames = []
for p in mask_dir.glob("frame_*.npz"):
try:
frames.append(int(p.stem.split("_")[1]))
except (ValueError, IndexError):
continue
return sorted(frames)
def resolve_frame_state(graph_json: dict, frame_idx: int) -> Tuple[Dict[str, bool], Dict[str, bool]]:
"""Resolve delta-encoded constraints + visibility at a frame.
Walks frame_states from frame 0 to frame_idx, accumulating deltas.
Returns (constraints_dict, visibility_dict).
"""
constraints: Dict[str, bool] = {}
visibility: Dict[str, bool] = {}
# Defaults: every component visible, every edge locked
for c in graph_json["components"]:
visibility[c["id"]] = True
for e in graph_json["edges"]:
constraints[f"{e['src']}->{e['dst']}"] = True
# Walk deltas up to frame_idx
fs_dict = graph_json.get("frame_states", {})
for f in sorted([int(k) for k in fs_dict]):
if f > frame_idx:
break
fs = fs_dict[str(f)]
for k, v in fs.get("constraints", {}).items():
constraints[k] = v
for k, v in fs.get("visibility", {}).items():
visibility[k] = v
return constraints, visibility
def type_one_hot(comp_type: str, type_vocab: List[str]) -> List[float]:
"""9-dim one-hot encoding of component type based on type_vocab."""
return [1.0 if t == comp_type else 0.0 for t in type_vocab]
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Raw data loader (NumPy only, no torch)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class FrameData:
graph: dict
masks: Dict[str, np.ndarray]
embeddings: Dict[str, np.ndarray]
depth_info: dict
robot: Optional[dict]
constraints: Dict[str, bool]
visibility: Dict[str, bool]
def load_frame_data(episode_dir: Path, frame_idx: int) -> FrameData:
"""Load all v3 annotation files for one frame."""
anno = episode_dir / "annotations"
with open(anno / "side_graph.json") as f:
graph = json.load(f)
def _load_npz_dict(path: Path) -> Dict[str, np.ndarray]:
if not path.exists():
return {}
d = np.load(path)
return {k: d[k] for k in d.files}
masks = _load_npz_dict(anno / "side_masks" / f"frame_{frame_idx:06d}.npz")
embeddings = _load_npz_dict(anno / "side_embeddings" / f"frame_{frame_idx:06d}.npz")
depth_info = _load_npz_dict(anno / "side_depth_info" / f"frame_{frame_idx:06d}.npz")
robot: Optional[dict] = None
robot_path = anno / "side_robot" / f"frame_{frame_idx:06d}.npz"
if robot_path.exists():
r = np.load(robot_path)
if r["visible"][0] == 1:
robot = {k: r[k] for k in r.files}
constraints, visibility = resolve_frame_state(graph, frame_idx)
return FrameData(graph, masks, embeddings, depth_info, robot, constraints, visibility)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# PyG loader β products only
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def load_pyg_frame_products_only(episode_dir: Path, frame_idx: int) -> Data:
"""Fully connected PyG graph WITHOUT robot.
Returns Data(
x=[N, 268],
edge_index=[2, N*(N-1)],
edge_attr=[N*(N-1), 3], # [has_constraint, is_locked, src_blocks_dst]
num_nodes=N,
)
where N = number of product components (robot excluded).
"""
fd = load_frame_data(episode_dir, frame_idx)
graph = fd.graph
type_vocab = graph["type_vocab"] # 9 entries incl. robot
nodes = graph["components"] # robot already excluded per spec
N = len(nodes)
# ββ Node features ββ
# [256D SAM2 embedding, 3D position, 9D type one-hot, 1D visibility] = 269
# NOTE: 256 + 3 + 9 + 1 = 269 (not 268). Adjust if you need a different layout.
x_list = []
for node in nodes:
cid = node["id"]
emb = fd.embeddings.get(cid, np.zeros(256, dtype=np.float32))
depth_valid_key = f"{cid}_depth_valid"
centroid_key = f"{cid}_centroid"
if (depth_valid_key in fd.depth_info
and int(fd.depth_info[depth_valid_key][0]) == 1):
pos = fd.depth_info[centroid_key].astype(np.float32)
else:
pos = np.zeros(3, dtype=np.float32)
type_oh = type_one_hot(node["type"], type_vocab) # 9D
vis = 1.0 if fd.visibility.get(cid, True) else 0.0
feat = np.concatenate([
emb.astype(np.float32),
pos,
np.array(type_oh, dtype=np.float32),
np.array([vis], dtype=np.float32),
])
x_list.append(feat)
x = torch.tensor(np.stack(x_list), dtype=torch.float32) if x_list else torch.empty((0, 269))
# ββ Fully connected edges with 3D features ββ
# Edge feature: [has_constraint, is_locked, src_blocks_dst]
# - has_constraint & is_locked are SYMMETRIC for the pair (A, B)
# - src_blocks_dst is ASYMMETRIC: 1 if edge's src physically blocks dst
constraint_set = {(e["src"], e["dst"]) for e in graph["edges"]}
pair_forward = {} # frozenset({a, b}) -> (blocker, blocked)
for (s, d) in constraint_set:
pair_forward[frozenset([s, d])] = (s, d)
src_idx, dst_idx, edge_attr = [], [], []
for i in range(N):
for j in range(N):
if i == j:
continue
src_id = nodes[i]["id"]
dst_id = nodes[j]["id"]
src_idx.append(i)
dst_idx.append(j)
pair_key = frozenset([src_id, dst_id])
if pair_key in pair_forward:
forward = pair_forward[pair_key]
constraint_key = f"{forward[0]}->{forward[1]}"
is_locked = fd.constraints.get(constraint_key, True)
src_blocks_dst = 1.0 if src_id == forward[0] else 0.0
edge_attr.append([
1.0,
1.0 if is_locked else 0.0,
src_blocks_dst,
])
else:
edge_attr.append([0.0, 0.0, 0.0]) # message passing only
return Data(
x=x,
edge_index=torch.tensor([src_idx, dst_idx], dtype=torch.long),
edge_attr=torch.tensor(edge_attr, dtype=torch.float32),
y=torch.tensor([frame_idx], dtype=torch.long),
num_nodes=N,
)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# PyG loader β with robot agent node
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def load_pyg_frame_with_robot(episode_dir: Path, frame_idx: int) -> Data:
"""Fully connected PyG graph WITH robot appended as agent node.
Robot is node N (the last node). All edges involving the robot have
features [0, 0, 0] because the robot has no physical constraints.
If the robot is not visible at this frame, returns the products-only graph.
Additional attached tensors when robot is visible:
data.robot_point_cloud (M, 3) float32
data.robot_pixel_coords (M, 2) int32
data.robot_mask (H, W) uint8
"""
data = load_pyg_frame_products_only(episode_dir, frame_idx)
fd = load_frame_data(episode_dir, frame_idx)
if fd.robot is None:
return data
graph = fd.graph
type_vocab = graph["type_vocab"]
products = graph["components"]
N_prod = len(products)
N = N_prod + 1
# ββ Build robot node features ββ
robot_emb = fd.robot["embedding"].astype(np.float32)
robot_pos = (fd.robot["centroid"].astype(np.float32)
if int(fd.robot["depth_valid"][0]) == 1
else np.zeros(3, dtype=np.float32))
robot_type_oh = type_one_hot("robot", type_vocab)
robot_feat = np.concatenate([
robot_emb, robot_pos,
np.array(robot_type_oh, dtype=np.float32),
np.array([1.0], dtype=np.float32),
])
x = torch.cat([data.x, torch.tensor(robot_feat, dtype=torch.float32).unsqueeze(0)], dim=0)
# ββ Rebuild edges with 3D features ββ
constraint_set = {(e["src"], e["dst"]) for e in graph["edges"]}
pair_forward = {}
for (s, d) in constraint_set:
pair_forward[frozenset([s, d])] = (s, d)
src_idx, dst_idx, edge_attr = [], [], []
# Products Γ Products
for i in range(N_prod):
for j in range(N_prod):
if i == j:
continue
src_id = products[i]["id"]
dst_id = products[j]["id"]
src_idx.append(i)
dst_idx.append(j)
pair_key = frozenset([src_id, dst_id])
if pair_key in pair_forward:
forward = pair_forward[pair_key]
is_locked = fd.constraints.get(f"{forward[0]}->{forward[1]}", True)
src_blocks_dst = 1.0 if src_id == forward[0] else 0.0
edge_attr.append([1.0, 1.0 if is_locked else 0.0, src_blocks_dst])
else:
edge_attr.append([0.0, 0.0, 0.0])
# Robot β Products (both directions, message-passing only)
robot_idx = N_prod
for i in range(N_prod):
src_idx.append(robot_idx); dst_idx.append(i); edge_attr.append([0.0, 0.0, 0.0])
src_idx.append(i); dst_idx.append(robot_idx); edge_attr.append([0.0, 0.0, 0.0])
data = Data(
x=x,
edge_index=torch.tensor([src_idx, dst_idx], dtype=torch.long),
edge_attr=torch.tensor(edge_attr, dtype=torch.float32),
y=torch.tensor([frame_idx], dtype=torch.long),
num_nodes=N,
)
data.robot_point_cloud = torch.tensor(fd.robot["point_cloud"], dtype=torch.float32)
data.robot_pixel_coords = torch.tensor(fd.robot["pixel_coords"], dtype=torch.int32)
data.robot_mask = torch.tensor(fd.robot["mask"], dtype=torch.uint8)
return data
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Episode iterator
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def iterate_episode(episode_dir: Path, with_robot: bool = True):
"""Yield (frame_idx, Data) pairs for all labeled frames in an episode."""
loader = load_pyg_frame_with_robot if with_robot else load_pyg_frame_products_only
for frame_idx in list_labeled_frames(episode_dir):
yield frame_idx, loader(episode_dir, frame_idx)
|