add_review_cleanup_to_colab

notebooks/colab_gas_valve_auto_label_train.py

@@ -481,11 +481,222 @@ else:
 
 # %% [markdown]
 # ---
-# ## Step 9 - Build the two-class training dataset
+# ## Step 9 - Review and cleaning controls
 
 
 # %%
-# @title Step 9 - Build the two-class training dataset
+# @title Step 9 - Review and cleaning controls
+# @markdown The manifest uses `source/stem` keys, for example `valve/Pic_阀门_60` or `gas_extra/Pic_气瓶_125`.
+# @markdown Put one key per line in `EXCLUDE_SAMPLE_KEYS` to remove those samples from the training dataset.
+
+import pandas as pd
+
+EXCLUDE_SAMPLE_KEYS = """
+"""  # @param {type:"string"}
+
+REVIEW_SOURCE = "valve"  # @param ["valve", "gas_extra", "old_gas", "all"]
+REVIEW_OFFSET = 0  # @param {type:"integer"}
+REVIEW_COUNT = 9  # @param {type:"integer"}
+
+DELETE_EXCLUDED_FROM_HF = False  # @param {type:"boolean"}
+CONFIRM_HF_DELETE = ""  # @param {type:"string"}
+
+
+def parse_exclude_keys(text: str) -> set[str]:
+    keys: set[str] = set()
+    for raw in text.replace(",", "\n").splitlines():
+        line = raw.strip()
+        if not line or line.startswith("#"):
+            continue
+        keys.add(line)
+    return keys
+
+
+EXCLUDED_SAMPLE_KEYS = parse_exclude_keys(EXCLUDE_SAMPLE_KEYS)
+print(f"Excluded sample keys: {len(EXCLUDED_SAMPLE_KEYS)}")
+for key in sorted(EXCLUDED_SAMPLE_KEYS):
+    print(f"  {key}")
+
+
+def label_summary(lines: list[str]) -> tuple[str, str]:
+    classes: list[str] = []
+    points: list[str] = []
+    for line in lines:
+        parts = line.split()
+        if not parts:
+            continue
+        classes.append(parts[0])
+        points.append(str((len(parts) - 1) // 2))
+    return ",".join(classes), ",".join(points)
+
+
+manifest_rows: list[dict] = []
+
+include_old_in_review = globals().get("INCLUDE_OLD_GAS_DATA", True)
+if include_old_in_review:
+    old_img_dir = base / "images/train"
+    old_lbl_dir = base / "labels/train"
+    for lbl_path in sorted(old_lbl_dir.glob("*.txt")):
+        img_path = find_image_by_stem(old_img_dir, lbl_path.stem)
+        lines = valid_seg_lines(lbl_path, {0})
+        classes, points = label_summary(lines)
+        manifest_rows.append({
+            "source_key": f"old_gas/{lbl_path.stem}",
+            "source": "old_gas",
+            "stem": lbl_path.stem,
+            "image_path": str(img_path) if img_path else "",
+            "status": "ok" if img_path and lines else "missing_image_or_label",
+            "classes": classes,
+            "points": points,
+        })
+
+for img_path in list_images(RAW_GAS_EXTRA_DIR):
+    lines = valid_seg_lines(GAS_EXTRA_LABEL_DIR / f"{img_path.stem}.txt", {0})
+    classes, points = label_summary(lines)
+    manifest_rows.append({
+        "source_key": f"gas_extra/{img_path.stem}",
+        "source": "gas_extra",
+        "stem": img_path.stem,
+        "image_path": str(img_path),
+        "status": "ok" if lines else "missing_gas_label",
+        "classes": classes,
+        "points": points,
+    })
+
+for img_path in list_images(RAW_VALVE_DIR):
+    gas_lines = valid_seg_lines(VALVE_GAS_LABEL_DIR / f"{img_path.stem}.txt", {0})
+    valve_lines = valid_seg_lines(VALVE_LABEL_DIR / f"{img_path.stem}.txt", {1})
+    classes, points = label_summary(gas_lines + valve_lines)
+    status = "ok" if valve_lines else "missing_valve_label"
+    manifest_rows.append({
+        "source_key": f"valve/{img_path.stem}",
+        "source": "valve",
+        "stem": img_path.stem,
+        "image_path": str(img_path),
+        "status": status,
+        "classes": classes,
+        "points": points,
+    })
+
+manifest_df = pd.DataFrame(manifest_rows)
+manifest_path = Path("/content/gas_valve_review_manifest.csv")
+manifest_df.to_csv(manifest_path, index=False, encoding="utf-8-sig")
+
+print("Review manifest written:", manifest_path)
+print(manifest_df.groupby(["source", "status"]).size().to_string())
+print("")
+print("First rows:")
+try:
+    display(manifest_df.head(20))
+except Exception:
+    print(manifest_df.head(20).to_string(index=False))
+
+
+def draw_lines_overlay(img_path: Path, lines: list[str]) -> np.ndarray:
+    img = read_image(img_path)
+    if img is None:
+        raise ValueError(f"Cannot read {img_path}")
+    h, w = img.shape[:2]
+    scale = 720 / max(w, h)
+    disp = cv2.resize(img, (int(w * scale), int(h * scale)), interpolation=cv2.INTER_AREA)
+    dh, dw = disp.shape[:2]
+    colors = {0: (0, 220, 0), 1: (255, 80, 0)}
+    for line in lines:
+        parts = line.split()
+        if len(parts) < 7:
+            continue
+        cls = int(parts[0])
+        vals = np.array([float(v) for v in parts[1:]], dtype=np.float32).reshape(-1, 2)
+        pts = (vals * np.array([dw, dh], dtype=np.float32)).astype(np.int32).reshape(-1, 1, 2)
+        color = colors.get(cls, (0, 255, 255))
+        overlay = disp.copy()
+        cv2.fillPoly(overlay, [pts], color)
+        cv2.addWeighted(overlay, 0.25, disp, 0.75, 0, disp)
+        cv2.polylines(disp, [pts], True, color, 2)
+        x, y = pts[0, 0]
+        cv2.putText(disp, str(cls), (int(x), int(y)), cv2.FONT_HERSHEY_SIMPLEX, 0.8, color, 2)
+    return cv2.cvtColor(disp, cv2.COLOR_BGR2RGB)
+
+
+review_df = manifest_df if REVIEW_SOURCE == "all" else manifest_df[manifest_df["source"] == REVIEW_SOURCE]
+review_df = review_df.iloc[max(0, REVIEW_OFFSET): max(0, REVIEW_OFFSET) + max(1, REVIEW_COUNT)]
+
+if len(review_df):
+    ncols = 3
+    nrows = int(np.ceil(len(review_df) / ncols))
+    fig, axes = plt.subplots(nrows, ncols, figsize=(18, 5 * nrows))
+    axes = np.array(axes).reshape(-1)
+    for ax, (_, row) in zip(axes, review_df.iterrows()):
+        source = row["source"]
+        stem = row["stem"]
+        img_path = Path(row["image_path"])
+        if source == "old_gas":
+            lines = valid_seg_lines(base / "labels/train" / f"{stem}.txt", {0})
+        elif source == "gas_extra":
+            lines = valid_seg_lines(GAS_EXTRA_LABEL_DIR / f"{stem}.txt", {0})
+        else:
+            lines = (
+                valid_seg_lines(VALVE_GAS_LABEL_DIR / f"{stem}.txt", {0}) +
+                valid_seg_lines(VALVE_LABEL_DIR / f"{stem}.txt", {1})
+            )
+        ax.imshow(draw_lines_overlay(img_path, lines))
+        ax.set_title(f"{row['source_key']}\\n{row['status']}", fontsize=9)
+        ax.axis("off")
+    for ax in axes[len(review_df):]:
+        ax.axis("off")
+    plt.tight_layout()
+    plt.show()
+else:
+    print("No samples selected for review.")
+
+
+if DELETE_EXCLUDED_FROM_HF:
+    if CONFIRM_HF_DELETE != "DELETE":
+        print("HF delete skipped: set CONFIRM_HF_DELETE to DELETE to enable deletion.")
+    else:
+        from huggingface_hub import CommitOperationDelete
+
+        remote_files = set(api.list_repo_files(DATASET_REPO, repo_type="dataset"))
+        delete_ops = []
+        for key in EXCLUDED_SAMPLE_KEYS:
+            if "/" not in key:
+                continue
+            source, stem = key.split("/", 1)
+            candidates: list[str] = []
+            if source == "valve":
+                candidates += [f for f in remote_files if f.startswith(f"{RAW_VALVE_PREFIX}/{stem}.")]
+                candidates += [
+                    f"{GENERATED_LABELS_REPO_DIR}/valve/{stem}.txt",
+                    f"{GENERATED_LABELS_REPO_DIR}/valve_gas/{stem}.txt",
+                ]
+            elif source == "gas_extra":
+                candidates += [f for f in remote_files if f.startswith(f"{RAW_GAS_EXTRA_PREFIX}/{stem}.")]
+                candidates += [f"{GENERATED_LABELS_REPO_DIR}/gas_extra/{stem}.txt"]
+            else:
+                print(f"Skipping HF delete for {key}; old_gas is the original dataset.")
+            for path in candidates:
+                if path in remote_files:
+                    delete_ops.append(CommitOperationDelete(path_in_repo=path))
+        if delete_ops:
+            api.create_commit(
+                repo_id=DATASET_REPO,
+                repo_type="dataset",
+                token=HF_TOKEN,
+                operations=delete_ops,
+                commit_message="delete_excluded_review_samples",
+            )
+            print(f"Deleted {len(delete_ops)} files from HF.")
+        else:
+            print("No HF files matched the excluded sample keys.")
+
+
+# %% [markdown]
+# ---
+# ## Step 10 - Build the two-class training dataset
+
+
+# %%
+# @title Step 10 - Build the two-class training dataset
 
 COMBINED_DATASET = Path("/content/gas_valve_dataset")
 VAL_RATIO = 0.20  # @param {type:"number"}
@@ -516,6 +727,8 @@ if INCLUDE_OLD_GAS_DATA:
             "lines": lines,
             "group": "gas",
             "source": "old_gas",
+            "source_stem": lbl_path.stem,
+            "source_key": f"old_gas/{lbl_path.stem}",
         })
 
 for img_path in list_images(RAW_GAS_EXTRA_DIR):
@@ -527,6 +740,8 @@ for img_path in list_images(RAW_GAS_EXTRA_DIR):
         "lines": lines,
         "group": "gas",
         "source": "gas_extra",
+        "source_stem": img_path.stem,
+        "source_key": f"gas_extra/{img_path.stem}",
     })
 
 for img_path in list_images(RAW_VALVE_DIR):
@@ -539,8 +754,19 @@ for img_path in list_images(RAW_VALVE_DIR):
         "lines": gas_lines + valve_lines,
         "group": "valve",
         "source": "valve",
+        "source_stem": img_path.stem,
+        "source_key": f"valve/{img_path.stem}",
     })
 
+excluded_keys = globals().get("EXCLUDED_SAMPLE_KEYS", set())
+if excluded_keys:
+    before_filter = len(samples)
+    samples = [
+        s for s in samples
+        if s["source_key"] not in excluded_keys and s["source_stem"] not in excluded_keys
+    ]
+    print(f"Excluded from training: {before_filter - len(samples)} samples")
+
 print(f"Collected labeled samples: {len(samples)}")
 print(f"  gas samples: {sum(1 for s in samples if s['group'] == 'gas')}")
 print(f"  valve samples: {sum(1 for s in samples if s['group'] == 'valve')}")
@@ -607,11 +833,11 @@ print(YAML_PATH.read_text(encoding="utf-8"))
 
 # %% [markdown]
 # ---
-# ## Step 10 - Visual sanity check
+# ## Step 11 - Visual sanity check
 
 
 # %%
-# @title Step 10 - Visual sanity check
+# @title Step 11 - Visual sanity check
 
 def draw_label_overlay(img_path: Path, label_path: Path) -> np.ndarray:
     img = read_image(img_path)
@@ -663,11 +889,11 @@ else:
 
 # %% [markdown]
 # ---
-# ## Step 11 - Train YOLO-seg
+# ## Step 12 - Train YOLO-seg
 
 
 # %%
-# @title Step 11 - Train YOLO-seg
+# @title Step 12 - Train YOLO-seg
 
 from ultralytics import YOLO
 
@@ -724,11 +950,11 @@ print(f"Best weights: {RUN_DIR / 'weights/best.pt'}")
 
 # %% [markdown]
 # ---
-# ## Step 12 - Review results
+# ## Step 13 - Review results
 
 
 # %%
-# @title Step 12 - Review results
+# @title Step 13 - Review results
 
 import pandas as pd
 import matplotlib.image as mpimg
@@ -755,11 +981,11 @@ for name in ["results.png", "confusion_matrix_normalized.png", "confusion_matrix
 
 # %% [markdown]
 # ---
-# ## Step 13 - Upload training results
+# ## Step 14 - Upload training results
 
 
 # %%
-# @title Step 13 - Upload training results
+# @title Step 14 - Upload training results
 
 UPLOAD_TRAINING_RESULTS = True  # @param {type:"boolean"}
 UPLOAD_REPO = DATASET_REPO  # @param {type:"string"}