oriented-det 0.1.0__py3-none-any.whl

oriented_det/cli/__init__.py

@@ -0,0 +1,92 @@
+"""``odet`` unified command-line interface."""
+
+from __future__ import annotations
+
+import importlib
+import sys
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+
+# Maps subcommand name -> (module path, argv[0] for the target script)
+_COMMANDS: Dict[str, Tuple[str, str]] = {
+    "train": ("tools.train", "odet-train"),
+    "train-multi-gpu": ("tools.train_multi_gpu", "odet-train-multi-gpu"),
+    "preds": ("tools.save_predictions", "odet-preds"),
+    "metrics": ("tools.save_predictions", "odet-metrics"),
+    "lr-finder": ("tools.lr_finder", "odet-lr-finder"),
+    "stats": ("tools.dataset_stats", "odet-stats"),
+    "tile-dota": ("tools.tile_dota", "odet-tile-dota"),
+    "image-demo": ("tools.image_demo", "odet-image-demo"),
+    "viewer": ("tools.app", "odet-viewer"),
+    "playground-csv": ("tools.generate_airbus_playground_csv", "odet-playground-csv"),
+    "playground-to-dota": ("tools.playground_to_dota", "odet-playground-to-dota"),
+    "export-onnx": ("export.scripts.export_onnx", "odet-export-onnx"),
+    "labels-to-comma": ("tools.dota_labels_to_comma", "odet-labels-to-comma"),
+    "free-gpu": ("tools.free_gpu", "odet-free-gpu"),
+    "visualize-boxes": ("tools.visualize_boxes", "odet-visualize-boxes"),
+    "pretrained": ("tools.pretrained_download", "odet-pretrained"),
+}
+
+_EXPORT_TF_ALIASES = frozenset({"export-tf", "export-detect", "export-preds"})
+
+
+def _print_help() -> None:
+    print("Usage: odet <command> [options]")
+    print("")
+    print("Commands:")
+    for name in sorted(_COMMANDS):
+        print(f"  {name}")
+    print("  pretrained      Download registered checkpoints from Hugging Face Hub")
+    print("  export-tf       Run export/Makefile export-tf (pass MAKE_ARGS='...')")
+    print("  export-onnx     (alias above) ONNX export via export/scripts/export_onnx.py")
+    print("")
+    print("Examples:")
+    print("  odet train --config configs/rotated_faster_rcnn/dota_le90_3x.json")
+    print("  odet train-multi-gpu --config configs/rotated_faster_rcnn/dota_le90_3x.json")
+    print("  odet preds --experiment-dir runs/rotated_faster_rcnn/<id>")
+    print("  odet playground-csv --data-root /path/to/export")
+
+
+def _invoke(module_path: str, prog: str, args: List[str]) -> None:
+    repo_root = Path(__file__).resolve().parents[2]
+    if str(repo_root) not in sys.path:
+        sys.path.insert(0, str(repo_root))
+    mod = importlib.import_module(module_path)
+    if not hasattr(mod, "main"):
+        raise SystemExit(f"{module_path} has no main()")
+    sys.argv = [prog] + args
+    mod.main()
+
+
+def _run_export_tf(args: List[str]) -> None:
+    import subprocess
+    from pathlib import Path
+
+    repo_root = Path(__file__).resolve().parents[2]
+    make_args = " ".join(args) if args else ""
+    cmd = f"make -C {repo_root / 'export'} export-tf {make_args}"
+    raise SystemExit(subprocess.call(cmd, shell=True))
+
+
+def main(argv: Optional[List[str]] = None) -> None:
+    argv = list(sys.argv[1:] if argv is None else argv)
+    if not argv or argv[0] in ("-h", "--help"):
+        _print_help()
+        return
+    cmd, rest = argv[0], argv[1:]
+    if cmd in ("-h", "--help"):
+        _print_help()
+        return
+    if cmd == "export-tf":
+        _run_export_tf(rest)
+        return
+    if cmd not in _COMMANDS:
+        print(f"Unknown command: {cmd}", file=sys.stderr)
+        _print_help()
+        raise SystemExit(2)
+    module_path, prog = _COMMANDS[cmd]
+    _invoke(module_path, prog, rest)
+
+
+if __name__ == "__main__":
+    main()

oriented_det/cli/train.py

@@ -0,0 +1,18 @@
+"""``odet train`` / ``torchrun -m oriented_det.cli.train`` entry point."""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+# tools/ lives at the oriented-det repo root (editable install); same bootstrap as oriented_det.cli._invoke
+_repo_root = Path(__file__).resolve().parents[2]
+if str(_repo_root) not in sys.path:
+    sys.path.insert(0, str(_repo_root))
+
+from tools.train import main
+
+__all__ = ["main"]
+
+if __name__ == "__main__":
+    main()

oriented_det/configs/_base_/augmentation.json

@@ -0,0 +1,21 @@
+{
+  "augmentation": {
+    "p_brightness_contrast": 0.5,
+    "brightness_limit": 0.2,
+    "contrast_limit": 0.2,
+    "p_gamma": 0.3,
+    "gamma_limit": [
+      80,
+      120
+    ],
+    "p_noise": 0.2,
+    "gauss_noise_var_limit": [
+      10.0,
+      50.0
+    ],
+    "p_blur": 0.2,
+    "blur_limit": 3,
+    "p_clahe": 0.3,
+    "clahe_clip_limit": 4.0
+  }
+}

oriented_det/configs/_base_/datasets/dota_le90.json

@@ -0,0 +1,21 @@
+{
+  "dataset": {
+    "data_root": "/path/to/data/DOTA-v1.0-tiled",
+    "train_tiles_dir": "/path/to/data/DOTA-v1.0-tiled/train",
+    "val_tiles_dir": "/path/to/data/DOTA-v1.0-tiled/val",
+    "overlap": 200,
+    "difficult_strategy": "keep"
+  },
+  "preprocessing": {
+    "normalize_mean": [
+      0.307941,
+      0.311986,
+      0.297761
+    ],
+    "normalize_std": [
+      0.192315,
+      0.187769,
+      0.181551
+    ]
+  }
+}

oriented_det/configs/_base_/fp16.json

@@ -0,0 +1,5 @@
+{
+  "training": {
+    "use_amp": true
+  }
+}

oriented_det/configs/_base_/models/oriented_rcnn_r50.json

@@ -0,0 +1,26 @@
+{
+  "model_type": "oriented_rcnn",
+  "model": {
+    "backbone": "resnet50",
+    "pretrained_backbone": true,
+    "fpn_returned_layers": [1, 2, 3, 4],
+    "fpn_strides": [4, 8, 16, 32, 64],
+    "trainable_layers": 5,
+    "anchor_scales": [8],
+    "anchor_ratios": [0.5, 1.0, 2.0],
+    "target_means": [0.0, 0.0, 0.0, 0.0, 0.0],
+    "target_stds": [0.1, 0.1, 0.2, 0.2, 0.1],
+    "roi_loss_type": "cross_entropy",
+    "roi_focal_alpha": 1.0,
+    "roi_focal_gamma": 2.0,
+    "roi_norm_factor": null,
+    "roi_edge_swap": true,
+    "roi_proj_xy": true,
+    "inference_pre_nms_score_threshold": 0.05,
+    "use_hbb_for_matching": true
+  },
+  "loss": {
+    "_muted_comment": "Plain ROI cross-entropy by default. Schedules might override with class_weighted or focal.",
+    "loss_type": "cross_entropy"
+  }
+}

oriented_det/configs/_base_/models/rotated_faster_rcnn_r50.json

@@ -0,0 +1,53 @@
+{
+  "model_type": "rotated_faster_rcnn",
+  "model": {
+    "backbone": "resnet50",
+    "pretrained_backbone": true,
+    "fpn_returned_layers": [1, 2, 3, 4],
+    "fpn_strides": [4, 8, 16, 32, 64],
+    "trainable_layers": 5,
+    "anchor_scales": [
+      8
+    ],
+    "anchor_ratios": [
+      0.5,
+      1.0,
+      2.0
+    ],
+    "target_means": [
+      0,
+      0,
+      0,
+      0,
+      0
+    ],
+    "target_stds": [
+      0.1,
+      0.1,
+      0.2,
+      0.2,
+      0.1
+    ],
+    "roi_loss_type": "cross_entropy",
+    "roi_focal_alpha": 1,
+    "roi_focal_gamma": 2,
+    "roi_norm_factor": 2,
+    "roi_edge_swap": true,
+    "roi_box_reg_angle_weight": 1.0,
+    "use_hbb_for_matching": true,
+    "inference_pre_nms_score_threshold": 0.1,
+    "rpn_min_size": 0,
+    "rpn_pre_nms_top_n": 5000,
+    "rpn_post_nms_top_n": 3000,
+    "rpn_batch_size_per_image": 256,
+    "rpn_nms_threshold": 0.7,
+    "final_nms_iou_threshold": 0.5,
+    "max_detections_per_image": 2000,
+    "roi_batch_size_per_image": 512,
+    "rpn_positive_iou_threshold": 0.7,
+    "rpn_negative_iou_threshold": 0.3,
+    "roi_positive_iou_threshold": 0.5,
+    "roi_negative_iou_threshold": 0.5,
+    "add_gt_as_proposals": true
+  }
+}

oriented_det/configs/_base_/models/rotated_retinanet_r50.json

@@ -0,0 +1,30 @@
+{
+  "model_type": "rotated_retinanet",
+  "model": {
+    "backbone": "resnet50",
+    "pretrained_backbone": true,
+    "trainable_layers": 5,
+    "anchor_scales": [8],
+    "anchor_ratios": [0.5, 1.0, 2.0],
+    "target_means": [0.0, 0.0, 0.0, 0.0, 0.0],
+    "target_stds": [1.0, 1.0, 1.0, 1.0, 1.0],
+    "roi_loss_type": "focal",
+    "roi_focal_alpha": 0.25,
+    "roi_focal_gamma": 2.0,
+    "roi_norm_factor": null,
+    "roi_edge_swap": true,
+    "use_hbb_for_matching": false,
+    "rpn_min_size": 2,
+    "rpn_pre_nms_top_n": 6000,
+    "rpn_post_nms_top_n": 3000,
+    "max_detections_per_image": 3000
+  },
+  "tensorboard": {
+    "log_debug_anchors_proposals": true
+  },
+  "loss": {
+    "loss_type": "focal",
+    "focal_alpha": 0.25,
+    "focal_gamma": 2.0
+  }
+}

oriented_det/configs/_base_/preprocessing.json

@@ -0,0 +1,8 @@
+{
+  "preprocessing": {
+    "pad_size_divisor": 32,
+    "enable_flip_horizontal": true,
+    "enable_flip_vertical": true,
+    "enable_flip_diagonal": false
+  }
+}

oriented_det/configs/_base_/schedules/1x.json

@@ -0,0 +1,33 @@
+{
+  "training": {
+    "num_epochs": 12,
+    "learning_rate": 0.02,
+    "momentum": 0.9,
+    "weight_decay": 0.0001,
+    "use_amp": false,
+    "gradient_accumulation_steps": 1,
+    "max_grad_norm": 5.0,
+    "lr_scheduler_step_epochs": 8,
+    "lr_scheduler_milestones": [8, 11],
+    "lr_scheduler_gamma": 0.1,
+    "lr_warmup_steps": 500,
+    "lr_scaling_with_accumulation": "linear",
+    "use_lr_param_groups": false
+    },
+  "evaluation": {
+    "score_threshold": 0.3,
+    "iou_threshold": 0.5,
+    "compute_map_final": true
+  },
+  "data_loader": {
+    "batch_size": 50,
+    "num_workers": 4,
+    "shuffle": true,
+    "pin_memory": true
+  },
+  "loss": {
+    "loss_type": "cross_entropy"
+  },
+  "enable_albumentation": false,
+  "enable_profiling": false
+}