着色破坏结构命令及其参数

fastcdm/core.py

@@ -1,4 +1,4 @@
-from fastcdm.render.render_worker import RenderWorker
+from fastcdm.render.render_worker import RenderResult, RenderWorker
 from fastcdm.matcher import update_inliers, HungarianMatcher, SimpleAffineTransform
 from fastcdm.clean import (
     clean,
@@ -291,17 +291,23 @@ def render(self, latex_list: list) -> list:
             latex_strings = [
                 f"$${s}$$" if not s.startswith("$$") else s for s in latex_list
             ]
-            imgs = self.render_worker.render(latex_strings)
+            results = self.render_worker.render(latex_strings)
         except Exception as e:
             print("Rendering failed:")
             print("=" * 30)
             print(traceback.format_exc())
             return []
 
-        assert len(imgs) == len(
+        assert len(results) == len(
             latex_strings
         ), "Number of rendered images must match number of input strings"
-        return imgs
+        return [result.image for result in results]
+
+    def render_results(self, latex_list: list) -> List[RenderResult]:
+        latex_strings = [
+            f"$${s}$$" if not s.startswith("$$") else s for s in latex_list
+        ]
+        return self.render_worker.render(latex_strings)
 
     def compute(self, gt: str, pred: str, visualize: bool = False) -> tuple:
         """

fastcdm/render/render_worker.py

@@ -2,7 +2,8 @@
 import cv2
 import random
 import numpy as np
-from typing import List
+from dataclasses import dataclass
+from typing import List, Optional
 
 from selenium import webdriver
 from selenium.webdriver.chrome.service import Service as ChromeService
@@ -13,6 +14,16 @@
 from webdriver_manager.chrome import ChromeDriverManager
 
 
+@dataclass
+class RenderResult:
+    image: Optional[np.ndarray]
+    error: bool
+    error_text: Optional[str]
+    width: int
+    height: int
+    error_type: Optional[str] = None
+
+
 class RenderWorker:
     """
     一个使用 Selenium Headless Chrome 渲染HTML内容的工具类。
@@ -76,7 +87,7 @@ def __init__(self, template_file: str, timeout: int = 15, driver_path: str = Non
             EC.presence_of_all_elements_located((By.ID, "container"))
         )
 
-    def render(self, contents: List[str]) -> List[np.ndarray]:
+    def render(self, contents: List[str]) -> List[RenderResult]:
         """
         渲染一组内容并返回每个元素的截图。
         """
@@ -113,13 +124,13 @@ def render(self, contents: List[str]) -> List[np.ndarray]:
 
         # 获取每个渲染元素的边界框
         rects = self.get_rects()
-        cropped_imgs = []
+        results = []
         img_h, img_w = fullpage_img.shape[:2]
 
         # 根据边界框裁剪出每个元素的图像
         for rect in rects:
             if rect is None:
-                cropped_imgs.append(None)
+                results.append(RenderResult(None, True, "Invalid capture rectangle", 0, 0, "invalid_capture"))
             else:
                 x, y, w, h = rect
                 # 计算一个小的随机边距，让截图更自然
@@ -132,9 +143,9 @@ def render(self, contents: List[str]) -> List[np.ndarray]:
                 y2 = min(img_h, y + h + border_size)
 
                 cropped = fullpage_img[y1:y2, x1:x2]
-                cropped_imgs.append(cropped)
+                results.append(RenderResult(cropped, cropped.size == 0, "Empty cropped image" if cropped.size == 0 else None, w, h, "empty_image" if cropped.size == 0 else None))
 
-        return cropped_imgs
+        return results
 
     def get_rects(self) -> list:
         """

fastcdm/tokenize.py

@@ -2,7 +2,10 @@
 import subprocess
 from pathlib import Path
 from typing import Tuple
-import sys
+
+
+class CDMPreprocessingError(RuntimeError):
+    """Raised when formula normalization fails or produces unusable output."""
 
 
 IMPLICIT_MULTIPLICATION_TARGETS = [
@@ -124,6 +127,7 @@
     "partial",
     "nabla",
     "int",
+    "limits",
     "oint",
     "sum",
     "prod",
@@ -198,9 +202,12 @@
 PATTERN_LATEX_CMD_CONCAT_CMD = re.compile(
     r"\\(" + TARGETS_PATTERN + r")" + r"(\\[a-zA-Z])"
 )
-PATTERN_LATEX_CMD_CONCAT_TEXT = re.compile(r"\\(" + TARGETS_PATTERN + r")([a-zA-Z])")
+PATTERN_LATEX_CMD_CONCAT_TEXT = re.compile(
+    r"\\(?!(?:" + TARGETS_PATTERN + r")(?![a-zA-Z]))"
+    r"(" + TARGETS_PATTERN + r")([a-zA-Z])"
+)
 PATTERN_NON_CMD_IMPLICIT_MULT = re.compile(
-    r"\b(" + TARGETS_PATTERN + r")([a-zA-Z][a-zA-Z0-9]*)\b"
+    r"(?<!\\)\b(" + TARGETS_PATTERN + r")([a-zA-Z][a-zA-Z0-9]*)\b"
 )
 
 OPERATORS = "\s?".join(
@@ -247,7 +254,27 @@
 
 
 
-def tokenize(latex_code: str) -> Tuple[bool, str]:
+def _validate_normalized(source: str, normalized: str) -> None:
+    if source.strip() and not normalized.strip():
+        raise CDMPreprocessingError("Tokenizer returned empty output for non-empty input")
+    if "[PROCESSING FAILED]" in normalized:
+        raise CDMPreprocessingError("Tokenizer returned a processing failure marker")
+    environments = []
+    for match in re.finditer(r"\\(begin|end)\s*\{\s*([^{}]*?)\s*\}", normalized):
+        # A backslash escaped by another backslash is not a command start.
+        preceding = len(normalized[:match.start()]) - len(normalized[:match.start()].rstrip("\\"))
+        if preceding % 2:
+            continue
+        command, environment = match.groups()
+        if command == "begin":
+            environments.append(environment)
+        elif not environments or environments.pop() != environment:
+            raise CDMPreprocessingError("Tokenizer returned mismatched begin/end environments")
+    if environments:
+        raise CDMPreprocessingError("Tokenizer returned mismatched begin/end environments")
+
+
+def tokenize(latex_code: str, timeout: float = 30.0) -> Tuple[bool, str]:
 
     if not latex_code:
         return True, ""
@@ -280,13 +307,14 @@ def tokenize(latex_code: str) -> Tuple[bool, str]:
             text=True,
             check=True,
             encoding="utf-8",
+            timeout=timeout,
         )
         normalized_latex = proc.stdout
-    except (subprocess.CalledProcessError, FileNotFoundError) as e:
-        print(f"执行 Node.js 脚本（公式）时出错：{e}", file=sys.stderr)
-        if hasattr(e, "stderr"):
-            print(f"Node.js stderr：{e.stderr}", file=sys.stderr)
-        return False, latex_code
+    except (subprocess.CalledProcessError, subprocess.TimeoutExpired, FileNotFoundError) as e:
+        detail = getattr(e, "stderr", None) or str(e)
+        raise CDMPreprocessingError(f"Node.js tokenizer failed: {detail}") from e
+
+    _validate_normalized(latex_code, normalized_latex)
 
     names = [
         "\\" + x.replace(" ", "")
@@ -295,4 +323,6 @@ def tokenize(latex_code: str) -> Tuple[bool, str]:
     post = PATTERN_OPERATOR_NAME.sub(
         lambda match: str(names.pop(0)), normalized_latex
     ).replace(r"\\ \end{array}", r"\end{array}")
-    return True, post.strip()
+    post = post.strip()
+    _validate_normalized(latex_code, post)
+    return True, post

fastcdm/tokenize_latex/preprocess_formula.js

@@ -62,7 +62,7 @@ rl.on('line', function(line){
         console.error(line);
         console.error(norm_str);
         console.error(e);
-        console.log();
+        process.exitCode = 1;
     }
     global_str = ""
     norm_str = ""
@@ -256,7 +256,7 @@ groupTypes.spacing = function(group) {
 groupTypes.op = function(group) {
     var node;
 
-    if (group.value.symbol) {
+    if (group.value.symbol || group.value.alwaysHandleSupSub) {
         // 直接输出符号
         norm_str = norm_str + group.value.body + " ";
 
@@ -271,6 +271,10 @@ groupTypes.op = function(group) {
         }
         norm_str = norm_str + "} ";
     }
+    // The parser sets this flag only for an explicit limit control.
+    if (group.value.alwaysHandleSupSub) {
+        norm_str += group.value.limits ? "\\limits " : "\\nolimits ";
+    }
 };
 
 groupTypes.katex = function(group) {