| |
| |
| """ |
| Benchmark Dataset Splitter & Renamer. |
| |
| This script reads the JSON mapping files generated by `trajectory_split_domain_aware.py`, |
| extracts the corresponding samples from the original merged trajectory data, and optionally |
| renames the output files with split-specific prefixes. |
| """ |
|
|
| from __future__ import annotations |
|
|
| import argparse |
| import json |
| from copy import deepcopy |
| from pathlib import Path |
| from typing import Dict, List |
|
|
| |
|
|
| DEFAULT_ORIGINAL_DATA_DIR = Path("Data/trajectories/trajectory_splits") |
| DEFAULT_MAPPING_DIR = Path("Data/trajectories/data_splits") |
| DEFAULT_OUTPUT_BASE = Path("Data/trajectories/data_splits") |
|
|
| DEFAULT_TRAIN_DIR = DEFAULT_OUTPUT_BASE / "train" |
| DEFAULT_VAL_DIR = DEFAULT_OUTPUT_BASE / "val" |
| DEFAULT_SCENE_UNSEEN_DIR = DEFAULT_OUTPUT_BASE / "test_scene_unseen" |
| DEFAULT_TRAJECTORY_UNSEEN_DIR = DEFAULT_OUTPUT_BASE / "test_trajectory_unseen" |
| DEFAULT_INSTRUCTION_UNSEEN_DIR = DEFAULT_OUTPUT_BASE / "test_instruction_unseen" |
|
|
| DEFAULT_PREFIXES = { |
| "train": "train_", |
| "val": "val_", |
| "scene_unseen": "test_", |
| "trajectory_unseen": "test_", |
| "instruction_unseen": "test_", |
| } |
|
|
| MAPPING_FILENAMES = { |
| "train": "GSNav-Bench_Train_Split_Domain.json", |
| "val": "GSNav-Bench_Val_Split_Domain.json", |
| "scene_unseen": "GSNav-Bench_Test_Scene_Unseen_Split_Domain.json", |
| "trajectory_unseen": "GSNav-Bench_Test_Trajectory_Unseen_Split_Domain.json", |
| "instruction_unseen": "GSNav-Bench_Test_Instruction_Unseen_Split_Domain.json", |
| } |
|
|
|
|
| |
|
|
|
|
| class BenchmarkDatasetBuilder: |
| """Extract split data according to mapping files and optionally rename outputs.""" |
|
|
| def __init__( |
| self, |
| original_data_dir: Path, |
| mapping_dir: Path, |
| output_dirs: Dict[str, Path], |
| prefixes: Dict[str, str], |
| rename_files: bool = True, |
| ): |
| self.original_data_dir = original_data_dir |
| self.mapping_dir = mapping_dir |
| self.output_dirs = output_dirs |
| self.prefixes = prefixes |
| self.rename_files = rename_files |
|
|
| self.split_mappings: Dict[str, Dict] = {} |
|
|
| |
|
|
| def load_split_mappings(self) -> None: |
| """Load all split mapping JSON files.""" |
| print("Loading split mapping files...") |
|
|
| for split_name, filename in MAPPING_FILENAMES.items(): |
| file_path = self.mapping_dir / filename |
| if not file_path.exists(): |
| raise FileNotFoundError(f"Mapping file not found: {file_path}") |
|
|
| with file_path.open("r", encoding="utf-8") as f: |
| self.split_mappings[split_name] = json.load(f) |
|
|
| scene_count = len(self.split_mappings[split_name].get("scenes", {})) |
| print(f" Loaded {split_name:<20s}: {scene_count} scenes") |
|
|
| def load_original_scene_data(self, scene_id: str) -> Dict: |
| """Load the original trajectory JSON for a scene.""" |
| scene_dir = self.original_data_dir / scene_id |
| trajectory_files = list(scene_dir.glob("trajectories_overall_*.json")) |
|
|
| if not trajectory_files: |
| raise FileNotFoundError(f"No trajectories_overall_*.json found for scene {scene_id}") |
|
|
| with trajectory_files[0].open("r", encoding="utf-8") as f: |
| return json.load(f) |
|
|
| |
|
|
| def create_output_directories(self) -> None: |
| """Create output directories for all splits.""" |
| print("Creating output directories...") |
| for split_name, output_path in self.output_dirs.items(): |
| output_path.mkdir(parents=True, exist_ok=True) |
| print(f" {split_name:<20s} -> {output_path}") |
|
|
| |
|
|
| def _get_scene_filename(self, scene_id: str) -> str: |
| scene_dir = self.original_data_dir / scene_id |
| trajectory_files = list(scene_dir.glob("trajectories_overall_*.json")) |
| if trajectory_files: |
| return trajectory_files[0].name |
| return f"trajectories_overall_{scene_id}.json" |
|
|
| def _save_scene_data(self, data: Dict, output_dir: Path, filename: str) -> None: |
| output_dir.mkdir(parents=True, exist_ok=True) |
| output_path = output_dir / filename |
| with output_path.open("w", encoding="utf-8") as f: |
| json.dump(data, f, indent=2, ensure_ascii=False) |
|
|
| |
|
|
| def extract_scene_unseen(self) -> None: |
| """Extract Scene-Unseen test data (full scenes).""" |
| print("\n=== Extracting Scene-Unseen test set ===") |
| split_data = self.split_mappings["scene_unseen"] |
| output_dir = self.output_dirs["scene_unseen"] |
|
|
| total_scenes = total_trajectories = total_instructions = 0 |
|
|
| for scene_id in split_data["scenes"].keys(): |
| print(f" Scene {scene_id}") |
| original_data = self.load_original_scene_data(scene_id) |
| scene_filename = self._get_scene_filename(scene_id) |
| scene_output_dir = output_dir / scene_id |
|
|
| self._save_scene_data(original_data, scene_output_dir, scene_filename) |
|
|
| samples = original_data["scenes"][0]["samples"] |
| total_scenes += 1 |
| total_trajectories += len(samples) |
| total_instructions += sum(len(sample["instructions"]) for sample in samples) |
|
|
| print( |
| f"Scene-Unseen done: {total_scenes} scenes, {total_trajectories} trajectories, {total_instructions} instructions" |
| ) |
|
|
| def extract_val(self) -> None: |
| """Extract validation data (full scenes).""" |
| print("\n=== Extracting validation set ===") |
| split_data = self.split_mappings["val"] |
| output_dir = self.output_dirs["val"] |
|
|
| total_scenes = total_trajectories = total_instructions = 0 |
|
|
| for scene_id in split_data["scenes"].keys(): |
| print(f" Scene {scene_id}") |
| original_data = self.load_original_scene_data(scene_id) |
| scene_filename = self._get_scene_filename(scene_id) |
| scene_output_dir = output_dir / scene_id |
|
|
| self._save_scene_data(original_data, scene_output_dir, scene_filename) |
|
|
| samples = original_data["scenes"][0]["samples"] |
| total_scenes += 1 |
| total_trajectories += len(samples) |
| total_instructions += sum(len(sample["instructions"]) for sample in samples) |
|
|
| print( |
| f"Validation done: {total_scenes} scenes, {total_trajectories} trajectories, {total_instructions} instructions" |
| ) |
|
|
| def extract_trajectory_unseen(self) -> None: |
| """Extract Trajectory-Unseen test data (subset of trajectories).""" |
| print("\n=== Extracting Trajectory-Unseen test set ===") |
| split_data = self.split_mappings["trajectory_unseen"] |
| output_dir = self.output_dirs["trajectory_unseen"] |
|
|
| total_scenes = total_trajectories = total_instructions = 0 |
|
|
| for scene_id, scene_info in split_data["scenes"].items(): |
| print(f" Scene {scene_id}") |
| original_data = self.load_original_scene_data(scene_id) |
| samples = original_data["scenes"][0]["samples"] |
| trajectory_map = {sample["trajectory_id"]: sample for sample in samples} |
|
|
| selected_samples = [] |
| scene_instruction_count = 0 |
|
|
| for traj_info in scene_info["trajectories"]: |
| traj_id = traj_info["trajectory_id"] |
| sample = trajectory_map.get(traj_id) |
| if sample: |
| selected_samples.append(deepcopy(sample)) |
| scene_instruction_count += len(sample["instructions"]) |
| else: |
| print(f" [WARN] Trajectory {traj_id} not found in scene {scene_id}") |
|
|
| if selected_samples: |
| new_data = deepcopy(original_data) |
| new_data["scenes"][0]["samples"] = selected_samples |
|
|
| scene_output_dir = output_dir / scene_id |
| scene_filename = self._get_scene_filename(scene_id) |
| self._save_scene_data(new_data, scene_output_dir, scene_filename) |
|
|
| total_scenes += 1 |
| total_trajectories += len(selected_samples) |
| total_instructions += scene_instruction_count |
|
|
| print( |
| f" Selected {len(selected_samples)} trajectories, {scene_instruction_count} instructions" |
| ) |
|
|
| print( |
| f"Trajectory-Unseen done: {total_scenes} scenes, {total_trajectories} trajectories, {total_instructions} instructions" |
| ) |
|
|
| def extract_instruction_unseen(self) -> None: |
| """Extract Instruction-Unseen test data (subset of instructions).""" |
| print("\n=== Extracting Instruction-Unseen test set ===") |
| split_data = self.split_mappings["instruction_unseen"] |
| output_dir = self.output_dirs["instruction_unseen"] |
|
|
| total_scenes = total_trajectories = total_instructions = 0 |
|
|
| for scene_id, scene_info in split_data["scenes"].items(): |
| print(f" Scene {scene_id}") |
| original_data = self.load_original_scene_data(scene_id) |
| samples = original_data["scenes"][0]["samples"] |
| trajectory_map = {sample["trajectory_id"]: sample for sample in samples} |
|
|
| selected_samples = [] |
| scene_instruction_count = 0 |
|
|
| for traj_info in scene_info["trajectories"]: |
| traj_id = traj_info["trajectory_id"] |
| indices = traj_info["selected_instruction_indices"] |
| sample = trajectory_map.get(traj_id) |
| if sample: |
| new_sample = deepcopy(sample) |
| new_sample["instructions"] = [ |
| sample["instructions"][idx] |
| for idx in indices |
| if 0 <= idx < len(sample["instructions"]) |
| ] |
| if new_sample["instructions"]: |
| selected_samples.append(new_sample) |
| scene_instruction_count += len(new_sample["instructions"]) |
| else: |
| print(f" [WARN] Trajectory {traj_id} not found in scene {scene_id}") |
|
|
| if selected_samples: |
| new_data = deepcopy(original_data) |
| new_data["scenes"][0]["samples"] = selected_samples |
|
|
| scene_output_dir = output_dir / scene_id |
| scene_filename = self._get_scene_filename(scene_id) |
| self._save_scene_data(new_data, scene_output_dir, scene_filename) |
|
|
| total_scenes += 1 |
| total_trajectories += len(selected_samples) |
| total_instructions += scene_instruction_count |
|
|
| print( |
| f" Selected {len(selected_samples)} trajectories, {scene_instruction_count} instructions" |
| ) |
|
|
| print( |
| f"Instruction-Unseen done: {total_scenes} scenes, {total_trajectories} trajectories, {total_instructions} instructions" |
| ) |
|
|
| def extract_train(self) -> None: |
| """Extract training data (exclude test trajectories/instructions).""" |
| print("\n=== Extracting training set ===") |
| split_data = self.split_mappings["train"] |
| output_dir = self.output_dirs["train"] |
|
|
| total_scenes = total_trajectories = total_instructions = 0 |
|
|
| for scene_id, scene_info in split_data["scenes"].items(): |
| print(f" Scene {scene_id}") |
| original_data = self.load_original_scene_data(scene_id) |
| samples = original_data["scenes"][0]["samples"] |
| trajectory_map = {sample["trajectory_id"]: sample for sample in samples} |
|
|
| selected_samples = [] |
| scene_instruction_count = 0 |
|
|
| for traj_info in scene_info["trajectories"]: |
| traj_id = traj_info["trajectory_id"] |
| available_indices = set(traj_info["available_instruction_indices"]) |
| sample = trajectory_map.get(traj_id) |
| if sample and available_indices: |
| new_sample = deepcopy(sample) |
| new_sample["instructions"] = [ |
| sample["instructions"][idx] |
| for idx in sorted(available_indices) |
| if 0 <= idx < len(sample["instructions"]) |
| ] |
| if new_sample["instructions"]: |
| selected_samples.append(new_sample) |
| scene_instruction_count += len(new_sample["instructions"]) |
| else: |
| print(f" [WARN] Trajectory {traj_id} not found or has no available instructions") |
|
|
| if selected_samples: |
| new_data = deepcopy(original_data) |
| new_data["scenes"][0]["samples"] = selected_samples |
|
|
| scene_output_dir = output_dir / scene_id |
| scene_filename = self._get_scene_filename(scene_id) |
| self._save_scene_data(new_data, scene_output_dir, scene_filename) |
|
|
| total_scenes += 1 |
| total_trajectories += len(selected_samples) |
| total_instructions += scene_instruction_count |
|
|
| print( |
| f" Selected {len(selected_samples)} trajectories, {scene_instruction_count} instructions" |
| ) |
|
|
| print( |
| f"Training set done: {total_scenes} scenes, {total_trajectories} trajectories, {total_instructions} instructions" |
| ) |
|
|
| |
|
|
| def rename_split_files(self) -> None: |
| """Rename files in each split directory using configured prefixes.""" |
| if not self.rename_files: |
| return |
|
|
| print("\n=== Renaming split files ===") |
| for split_name, directory_path in self.output_dirs.items(): |
| prefix = self.prefixes.get(split_name) |
| if not prefix: |
| continue |
| self._rename_files_in_directory(directory_path, prefix, split_name) |
| self._verify_renaming() |
|
|
| def _rename_files_in_directory(self, directory_path: Path, prefix: str, split_name: str) -> None: |
| if not directory_path.exists(): |
| print(f"[WARN] Directory does not exist: {directory_path}") |
| return |
|
|
| total_files = renamed_files = 0 |
|
|
| for scene_dir in directory_path.iterdir(): |
| if scene_dir.is_dir(): |
| for file_path in scene_dir.iterdir(): |
| if file_path.is_file(): |
| total_files += 1 |
| if not file_path.name.startswith(prefix): |
| new_path = scene_dir / f"{prefix}{file_path.name}" |
| file_path.rename(new_path) |
| renamed_files += 1 |
|
|
| print( |
| f" {split_name:<20s}: {renamed_files}/{total_files} files renamed with prefix '{prefix}'" |
| ) |
|
|
| def _verify_renaming(self) -> None: |
| print("\n=== Rename verification ===") |
| for split_name, directory_path in self.output_dirs.items(): |
| if not directory_path.exists(): |
| print(f"{split_name:<20s}: directory missing") |
| continue |
|
|
| prefix = self.prefixes.get(split_name, "") |
| total_files = prefixed_files = 0 |
|
|
| for scene_dir in directory_path.iterdir(): |
| if scene_dir.is_dir(): |
| for file_path in scene_dir.iterdir(): |
| if file_path.is_file(): |
| total_files += 1 |
| if file_path.name.startswith(prefix): |
| prefixed_files += 1 |
|
|
| if total_files > 0: |
| ratio = prefixed_files / total_files * 100 |
| print(f"{split_name:<20s}: {prefixed_files}/{total_files} ({ratio:.1f}% prefixed)") |
| else: |
| print(f"{split_name:<20s}: no files") |
|
|
| |
|
|
| def run(self) -> None: |
| """Run the full pipeline: load mappings, extract splits, rename.""" |
| print("=" * 80) |
| print(" Benchmark Dataset Split & Rename") |
| print("=" * 80) |
|
|
| self.load_split_mappings() |
| self.create_output_directories() |
|
|
| self.extract_scene_unseen() |
| self.extract_trajectory_unseen() |
| self.extract_instruction_unseen() |
| self.extract_train() |
| self.extract_val() |
|
|
| self.rename_split_files() |
|
|
| print("\nDone. Output directories:") |
| for split_name, path in self.output_dirs.items(): |
| scene_count = len([d for d in path.iterdir() if d.is_dir()]) if path.exists() else 0 |
| print(f" {split_name:<20s}: {path} ({scene_count} scenes)") |
|
|
|
|
| |
|
|
|
|
| def parse_args() -> argparse.Namespace: |
| parser = argparse.ArgumentParser(description="Extract benchmark splits and rename files.") |
| parser.add_argument( |
| "--original-data-dir", |
| type=Path, |
| default=DEFAULT_ORIGINAL_DATA_DIR, |
| help="Directory containing original merged trajectory scenes", |
| ) |
| parser.add_argument( |
| "--mapping-dir", |
| type=Path, |
| default=DEFAULT_MAPPING_DIR, |
| help="Directory containing split mapping JSON files", |
| ) |
| parser.add_argument( |
| "--train-dir", |
| type=Path, |
| default=DEFAULT_TRAIN_DIR, |
| help="Output directory for the training split", |
| ) |
| parser.add_argument( |
| "--val-dir", |
| type=Path, |
| default=DEFAULT_VAL_DIR, |
| help="Output directory for the validation split", |
| ) |
| parser.add_argument( |
| "--scene-unseen-dir", |
| type=Path, |
| default=DEFAULT_SCENE_UNSEEN_DIR, |
| help="Output directory for the scene-unseen split", |
| ) |
| parser.add_argument( |
| "--trajectory-unseen-dir", |
| type=Path, |
| default=DEFAULT_TRAJECTORY_UNSEEN_DIR, |
| help="Output directory for the trajectory-unseen split", |
| ) |
| parser.add_argument( |
| "--instruction-unseen-dir", |
| type=Path, |
| default=DEFAULT_INSTRUCTION_UNSEEN_DIR, |
| help="Output directory for the instruction-unseen split", |
| ) |
| parser.add_argument( |
| "--no-rename", |
| action="store_true", |
| help="Disable renaming of output files with split-specific prefixes", |
| ) |
|
|
| return parser.parse_args() |
|
|
|
|
| def main() -> None: |
| args = parse_args() |
|
|
| |
| if not args.original_data_dir.exists(): |
| print(f"[ERROR] Original data dir does not exist: {args.original_data_dir}") |
| return |
| if not args.mapping_dir.exists(): |
| print(f"[ERROR] Mapping dir does not exist: {args.mapping_dir}") |
| return |
|
|
| output_dirs = { |
| "train": args.train_dir, |
| "val": args.val_dir, |
| "scene_unseen": args.scene_unseen_dir, |
| "trajectory_unseen": args.trajectory_unseen_dir, |
| "instruction_unseen": args.instruction_unseen_dir, |
| } |
|
|
| builder = BenchmarkDatasetBuilder( |
| original_data_dir=args.original_data_dir, |
| mapping_dir=args.mapping_dir, |
| output_dirs=output_dirs, |
| prefixes=DEFAULT_PREFIXES, |
| rename_files=not args.no_rename, |
| ) |
| builder.run() |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
