Update mineru/model/mfr/unimernet/Unimernet.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

.github/workflows/cla.yml

@@ -18,9 +18,9 @@ jobs:
     steps:
       - name: "CLA Assistant"
         if: (github.event.comment.body == 'recheck' || github.event.comment.body == 'I have read the CLA Document and I hereby sign the CLA') || github.event_name == 'pull_request_target'
-        uses: contributor-assistant/github-action@v2.5.0
+        uses: contributor-assistant/github-action@v2.6.1
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN  }}
           # the below token should have repo scope and must be manually added by you in the repository's secret
           # This token is required only if you have configured to store the signatures in a remote repository/organization
           PERSONAL_ACCESS_TOKEN: ${{ secrets.RELEASE_TOKEN }}
@@ -28,7 +28,7 @@ jobs:
           path-to-signatures: 'signatures/version1/cla.json'
           path-to-document: 'https://github.com/opendatalab/MinerU/blob/master/MinerU_CLA.md' # e.g. a CLA or a DCO document
           # branch should not be protected
-          branch: 'master'
+          branch: 'cla'
           allowlist: myhloli,dt-yy,Focusshang,renpengli01,icecraft,drunkpig,wangbinDL,qiangqiang199,GDDGCZ518,papayalove,conghui,quyuan,LollipopsAndWine,Sidney233
 
          # the followings are the optional inputs - If the optional inputs are not given, then default values will be taken

.github/workflows/cli.yml

@@ -1,16 +1,15 @@
 # This workflow will install Python dependencies, run tests and lint with a variety of Python versions
 # For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python
 
-name: mineru
+name: mineru-cli-test
 on:
-  pull_request:
+  push:
     branches:
       - "master"
       - "dev"
     paths-ignore:
       - "cmds/**"
       - "**.md"
-  workflow_dispatch:
 jobs:
   cli-test:
     if: github.repository == 'opendatalab/MinerU'
@@ -20,31 +19,30 @@ jobs:
       fail-fast: true
 
     steps:
-    - name: PDF cli
-      uses: actions/checkout@v4
-      with:
-        ref: dev
-        fetch-depth: 2
+      - name: PDF cli
+        uses: actions/checkout@v4
+        with:
+          ref: dev
+          fetch-depth: 2
 
-    - name: install uv
-      uses: astral-sh/setup-uv@v5
-
-    - name: install&test
-      run: |
-        uv --version
-        uv venv --python 3.12
-        source .venv/bin/activate
-        uv pip install .[test]
-        cd $GITHUB_WORKSPACE && python tests/clean_coverage.py      
-        cd $GITHUB_WORKSPACE && coverage run
-        cd $GITHUB_WORKSPACE && python tests/get_coverage.py
+      - name: install uv
+        uses: astral-sh/setup-uv@v5
 
+      - name: install&test
+        run: |
+          uv --version
+          uv venv --python 3.12
+          source .venv/bin/activate
+          uv pip install .[test]
+          cd $GITHUB_WORKSPACE && python tests/clean_coverage.py      
+          cd $GITHUB_WORKSPACE && coverage run
+          cd $GITHUB_WORKSPACE && python tests/get_coverage.py
 
   notify_to_feishu:
     if: ${{ always() && !cancelled() && contains(needs.*.result, 'failure')}}
     needs: cli-test
     runs-on: ubuntu-latest
     steps:
-    - name: notify
-      run: |
-        curl -X POST -H "Content-Type: application/json" -d '{"msg_type":"post","content":{"post":{"zh_cn":{"title":"'${{ github.repository }}' GitHubAction Failed","content":[[{"tag":"text","text":""},{"tag":"a","text":"Please click here for details ","href":"https://github.com/'${{ github.repository }}'/actions/runs/'${GITHUB_RUN_ID}'"}]]}}}}'  ${{ secrets.FEISHU_WEBHOOK_URL }}
+      - name: notify
+        run: |
+          curl -X POST -H "Content-Type: application/json" -d '{"msg_type":"post","content":{"post":{"zh_cn":{"title":"'${{ github.repository }}' GitHubAction Failed","content":[[{"tag":"text","text":""},{"tag":"a","text":"Please click here for details ","href":"https://github.com/'${{ github.repository }}'/actions/runs/'${GITHUB_RUN_ID}'"}]]}}}}'  ${{ secrets.FEISHU_WEBHOOK_URL }}

.github/workflows/huigui.yml

@@ -1,48 +0,0 @@
-# This workflow will install Python dependencies, run tests and lint with a variety of Python versions
-# For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python
-
-name: mineru
-on:
-  push:
-    branches:
-      - "master"
-      - "dev"
-    paths-ignore:
-      - "cmds/**"
-      - "**.md"
-jobs:
-  cli-test:
-    if: github.repository == 'opendatalab/MinerU'
-    runs-on: ubuntu-latest
-    timeout-minutes: 240
-    strategy:
-      fail-fast: true
-
-    steps:
-    - name: PDF cli
-      uses: actions/checkout@v4
-      with:
-        ref: dev
-        fetch-depth: 2
-
-    - name: install uv
-      uses: astral-sh/setup-uv@v5
-
-    - name: install&test
-      run: |
-        uv --version
-        uv venv --python 3.12
-        source .venv/bin/activate
-        uv pip install .[test]
-        cd $GITHUB_WORKSPACE && python tests/clean_coverage.py      
-        cd $GITHUB_WORKSPACE && coverage run
-        cd $GITHUB_WORKSPACE && python tests/get_coverage.py
-
-  notify_to_feishu:
-    if: ${{ always() && !cancelled() && contains(needs.*.result, 'failure')}}
-    needs: cli-test
-    runs-on: ubuntu-latest
-    steps:
-    - name: notify
-      run: |
-        curl -X POST -H "Content-Type: application/json" -d '{"msg_type":"post","content":{"post":{"zh_cn":{"title":"'${{ github.repository }}' GitHubAction Failed","content":[[{"tag":"text","text":""},{"tag":"a","text":"Please click here for details ","href":"https://github.com/'${{ github.repository }}'/actions/runs/'${GITHUB_RUN_ID}'"}]]}}}}'  ${{ secrets.FEISHU_WEBHOOK_URL }}

README.md

@@ -43,7 +43,10 @@
 </div>
 
 # Changelog
-
+- 2025/07/22 2.1.3 Released
+  - Bug Fixes
+    - Fixed the issue of excessive memory consumption during the `MFR` step in the `pipeline` backend under certain scenarios #2771
+    - Fixed the inaccurate matching between `image`/`table` and `caption`/`footnote` under certain conditions #3129
 - 2025/07/16 2.1.1 Released
   - Bug fixes
     - Fixed text block content loss issue that could occur in certain `pipeline` scenarios #3005
@@ -59,10 +62,10 @@
     - Greatly enhanced post-processing speed when the `pipeline` backend handles batch processing of documents with fewer pages (<10 pages).
     - Layout analysis speed of the `pipeline` backend has been increased by approximately 20%.
   - **Experience Enhancements:**
-    - Built-in ready-to-use `fastapi service` and `gradio webui`. For detailed usage instructions, please refer to [Documentation](#3-api-calls-or-visual-invocation).
+    - Built-in ready-to-use `fastapi service` and `gradio webui`. For detailed usage instructions, please refer to [Documentation](https://opendatalab.github.io/MinerU/usage/quick_usage/#advanced-usage-via-api-webui-sglang-clientserver).
     - Adapted to `sglang` version `0.4.8`, significantly reducing the GPU memory requirements for the `vlm-sglang` backend. It can now run on graphics cards with as little as `8GB GPU memory` (Turing architecture or newer).
     - Added transparent parameter passing for all commands related to `sglang`, allowing the `sglang-engine` backend to receive all `sglang` parameters consistently with the `sglang-server`.
-    - Supports feature extensions based on configuration files, including `custom formula delimiters`, `enabling heading classification`, and `customizing local model directories`. For detailed usage instructions, please refer to [Documentation](#4-extending-mineru-functionality-through-configuration-files).
+    - Supports feature extensions based on configuration files, including `custom formula delimiters`, `enabling heading classification`, and `customizing local model directories`. For detailed usage instructions, please refer to [Documentation](https://opendatalab.github.io/MinerU/usage/quick_usage/#extending-mineru-functionality-with-configuration-files).
   - **New Features:**
     - Updated the `pipeline` backend with the PP-OCRv5 multilingual text recognition model, supporting text recognition in 37 languages such as French, Spanish, Portuguese, Russian, and Korean, with an average accuracy improvement of over 30%. [Details](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html)
     - Introduced limited support for vertical text layout in the `pipeline` backend.

README_zh-CN.md

@@ -43,6 +43,10 @@
 </div>
 
 # 更新记录
+- 2025/07/22 2.1.3发布
+  - bug修复
+    - 修复`pipeline`后端中`MFR`步骤在某些情况下显存消耗过大的问题 #2771
+    - 修复某些情况下`image`/`table`与`caption`/`footnote`匹配不准确的问题 #3129
 - 2025/07/16 2.1.1发布
   - bug修复 
     - 修复`pipeline`在某些情况可能发生的文本块内容丢失问题 #3005
@@ -56,12 +60,12 @@
   - 性能优化： 
     - 大幅提升某些特定分辨率（长边2000像素左右）文档的预处理速度
     - 大幅提升`pipeline`后端批量处理大量页数较少（<10）文档时的后处理速度
-    - `pipline`后端的layout分析速度提升约20%
+    - `pipeline`后端的layout分析速度提升约20%
   - 体验优化：
-    - 内置开箱即用的`fastapi服务`和`gradio webui`，详细使用方法请参考[文档](#3-api-调用-或-可视化调用)
+    - 内置开箱即用的`fastapi服务`和`gradio webui`，详细使用方法请参考[文档](https://opendatalab.github.io/MinerU/zh/usage/quick_usage/#apiwebuisglang-clientserver)
     - `sglang`适配`0.4.8`版本，大幅降低`vlm-sglang`后端的显存要求，最低可在`8G显存`(Turing及以后架构)的显卡上运行
     - 对所有命令增加`sglang`的参数透传，使得`sglang-engine`后端可以与`sglang-server`一致，接收`sglang`的所有参数
-    - 支持基于配置文件的功能扩展，包含`自定义公式标识符`、`开启标题分级功能`、`自定义本地模型目录`，详细使用方法请参考[文档](#4-基于配置文件扩展-mineru-功能)
+    - 支持基于配置文件的功能扩展，包含`自定义公式标识符`、`开启标题分级功能`、`自定义本地模型目录`，详细使用方法请参考[文档](https://opendatalab.github.io/MinerU/zh/usage/quick_usage/#mineru_1)
   - 新特性：  
     - `pipeline`后端更新 PP-OCRv5 多语种文本识别模型，支持法语、西班牙语、葡萄牙语、俄语、韩语等 37 种语言的文字识别，平均精度涨幅超30%。[详情](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html)
     - `pipeline`后端增加对竖排文本的有限支持

docker/china/Dockerfile

@@ -3,14 +3,18 @@ FROM lmsysorg/sglang:v0.4.8.post1-cu126
 
 # Install libgl for opencv support & Noto fonts for Chinese characters
 RUN apt-get update && \
-    apt-get install -y fonts-noto-core fonts-noto-cjk && \
-    apt-get install -y libgl1 && \
-    apt-get clean && \
+    apt-get install -y \
+        fonts-noto-core \
+        fonts-noto-cjk \
+        fontconfig \
+        libgl1 && \
     fc-cache -fv && \
+    apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
 # Install mineru latest
-RUN python3 -m pip install -U 'mineru[core]' -i https://mirrors.aliyun.com/pypi/simple --break-system-packages
+RUN python3 -m pip install -U 'mineru[core]' -i https://mirrors.aliyun.com/pypi/simple --break-system-packages && \
+    python3 -m pip cache purge
 
 # Download models and update the configuration file
 RUN /bin/bash -c "mineru-models-download -s modelscope -m all"

docker/global/Dockerfile

@@ -1,16 +1,20 @@
 # Use the official sglang image
 FROM lmsysorg/sglang:v0.4.8.post1-cu126
 
-# Install libgl for opencv support
+# Install libgl for opencv support & Noto fonts for Chinese characters
 RUN apt-get update && \
-    apt-get install -y fonts-noto-core fonts-noto-cjk && \
-    apt-get install -y libgl1 && \
-    apt-get clean && \
+    apt-get install -y \
+        fonts-noto-core \
+        fonts-noto-cjk \
+        fontconfig \
+        libgl1 && \
     fc-cache -fv && \
+    apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
 # Install mineru latest
-RUN python3 -m pip install -U 'mineru[core]' --break-system-packages
+RUN python3 -m pip install -U 'mineru[core]' --break-system-packages && \
+    python3 -m pip cache purge
 
 # Download models and update the configuration file
 RUN /bin/bash -c "mineru-models-download -s huggingface -m all"

docs/en/quick_start/docker_deployment.md

@@ -26,7 +26,7 @@ MinerU's Docker uses `lmsysorg/sglang` as the base image, so it includes the `sg
 >
 > If your device doesn't meet the above requirements, you can still use other features of MinerU, but cannot use `sglang` to accelerate VLM model inference, meaning you cannot use the `vlm-sglang-engine` backend or start the `vlm-sglang-server` service.
 
-## Start Docker Container:
+## Start Docker Container
 
 ```bash
 docker run --gpus all \
@@ -60,7 +60,7 @@ wget https://gcore.jsdelivr.net/gh/opendatalab/MinerU@master/docker/compose.yaml
 ### Start sglang-server service
 connect to `sglang-server` via `vlm-sglang-client` backend
   ```bash
-  docker compose -f compose.yaml --profile mineru-sglang-server up -d
+  docker compose -f compose.yaml --profile sglang-server up -d
   ```
   >[!TIP]
   >In another terminal, connect to sglang server via sglang client (only requires CPU and network, no sglang environment needed)
@@ -72,7 +72,7 @@ connect to `sglang-server` via `vlm-sglang-client` backend
 
 ### Start Web API service
   ```bash
-  docker compose -f compose.yaml --profile mineru-api up -d
+  docker compose -f compose.yaml --profile api up -d
   ```
   >[!TIP]
   >Access `http://<server_ip>:8000/docs` in your browser to view the API documentation.
@@ -81,7 +81,7 @@ connect to `sglang-server` via `vlm-sglang-client` backend
 
 ### Start Gradio WebUI service
   ```bash
-  docker compose -f compose.yaml --profile mineru-gradio up -d
+  docker compose -f compose.yaml --profile gradio up -d
   ```
   >[!TIP]
   >

docs/zh/quick_start/docker_deployment.md

@@ -25,7 +25,7 @@ Mineru的docker使用了`lmsysorg/sglang`作为基础镜像，因此在docker中
 >
 > 如果您的设备不满足上述条件，您仍然可以使用MinerU的其他功能，但无法使用`sglang`加速VLM模型推理，即无法使用`vlm-sglang-engine`后端和启动`vlm-sglang-server`服务。
 
-## 启动 Docker 容器：
+## 启动 Docker 容器
 
 ```bash
 docker run --gpus all \
@@ -58,7 +58,7 @@ wget https://gcore.jsdelivr.net/gh/opendatalab/MinerU@master/docker/compose.yaml
 ### 启动 sglang-server 服务
 并通过`vlm-sglang-client`后端连接`sglang-server`
   ```bash
-  docker compose -f compose.yaml --profile mineru-sglang-server up -d
+  docker compose -f compose.yaml --profile sglang-server up -d
   ```
   >[!TIP]
   >在另一个终端中通过sglang client连接sglang server（只需cpu与网络，不需要sglang环境）
@@ -70,7 +70,7 @@ wget https://gcore.jsdelivr.net/gh/opendatalab/MinerU@master/docker/compose.yaml
 
 ### 启动 Web API 服务
   ```bash
-  docker compose -f compose.yaml --profile mineru-api up -d
+  docker compose -f compose.yaml --profile api up -d
   ```
   >[!TIP]
   >在浏览器中访问 `http://<server_ip>:8000/docs` 查看API文档。
@@ -79,7 +79,7 @@ wget https://gcore.jsdelivr.net/gh/opendatalab/MinerU@master/docker/compose.yaml
 
 ### 启动 Gradio WebUI 服务
   ```bash
-  docker compose -f compose.yaml --profile mineru-gradio up -d
+  docker compose -f compose.yaml --profile gradio up -d
   ```
   >[!TIP]
   > 

mineru/backend/pipeline/batch_analyze.py

@@ -12,6 +12,7 @@ from ...utils.ocr_utils import get_adjusted_mfdetrec_res, get_ocr_result_list, O
 YOLO_LAYOUT_BASE_BATCH_SIZE = 8
 MFD_BASE_BATCH_SIZE = 1
 MFR_BASE_BATCH_SIZE = 16
+OCR_DET_BASE_BATCH_SIZE = 16
 
 
 class BatchAnalyze:
@@ -170,9 +171,9 @@ class BatchAnalyze:
                         batch_images.append(padded_img)
 
                     # 批处理检测
-                    batch_size = min(len(batch_images), self.batch_ratio * 16)  # 增加批处理大小
-                    # logger.debug(f"OCR-det batch: {batch_size} images, target size: {target_h}x{target_w}")
-                    batch_results = ocr_model.text_detector.batch_predict(batch_images, batch_size)
+                    det_batch_size = min(len(batch_images), self.batch_ratio * OCR_DET_BASE_BATCH_SIZE)  # 增加批处理大小
+                    # logger.debug(f"OCR-det batch: {det_batch_size} images, target size: {target_h}x{target_w}")
+                    batch_results = ocr_model.text_detector.batch_predict(batch_images, det_batch_size)
 
                     # 处理批处理结果
                     for i, (crop_info, (dt_boxes, elapse)) in enumerate(zip(group_crops, batch_results)):

mineru/backend/pipeline/pipeline_analyze.py

@@ -74,10 +74,10 @@ def doc_analyze(
         table_enable=True,
 ):
     """
-    适当调大MIN_BATCH_INFERENCE_SIZE可以提高性能，可能会增加显存使用量，
-    可通过环境变量MINERU_MIN_BATCH_INFERENCE_SIZE设置，默认值为128。
+    适当调大MIN_BATCH_INFERENCE_SIZE可以提高性能，更大的 MIN_BATCH_INFERENCE_SIZE会消耗更多内存，
+    可通过环境变量MINERU_MIN_BATCH_INFERENCE_SIZE设置，默认值为384。
     """
-    min_batch_inference_size = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 128))
+    min_batch_inference_size = int(os.environ.get('MINERU_MIN_BATCH_INFERENCE_SIZE', 384))
 
     # 收集所有页面信息
     all_pages_info = []  # 存储(dataset_index, page_index, img, ocr, lang, width, height)

mineru/backend/pipeline/pipeline_magic_model.py

@@ -275,7 +275,8 @@ class MagicModel:
 
 
             fst_idx, fst_kind, left_x, top_y = candidates[0]
-            candidates.sort(key=lambda x: (x[2] - left_x) ** 2 + (x[3] - top_y)**2)
+            fst_bbox = subjects[fst_idx]['bbox'] if fst_kind == SUB_BIT_KIND else objects[fst_idx - OBJ_IDX_OFFSET]['bbox']
+            candidates.sort(key=lambda x: bbox_distance(fst_bbox, subjects[x[0]]['bbox']) if x[1] == SUB_BIT_KIND else bbox_distance(fst_bbox, objects[x[0] - OBJ_IDX_OFFSET]['bbox']))
             nxt = None
 
             for i in range(1, len(candidates)):
@@ -294,7 +295,8 @@ class MagicModel:
             pair_dis = bbox_distance(subjects[sub_idx]['bbox'], objects[obj_idx]['bbox'])
             nearest_dis = float('inf')
             for i in range(N):
-                if i in seen_idx or i == sub_idx:continue
+                # 取消原先算法中 1对1 匹配的偏置
+                # if i in seen_idx or i == sub_idx:continue
                 nearest_dis = min(nearest_dis, bbox_distance(subjects[i]['bbox'], objects[obj_idx]['bbox']))
 
             if pair_dis >= 3*nearest_dis:

mineru/model/mfr/unimernet/Unimernet.py

@@ -104,6 +104,10 @@ class UnimernetModel(object):
 
         # Create dataset with sorted images
         dataset = MathDataset(sorted_images, transform=self.model.transform)
+
+        # 如果batch_size> len(sorted_images)，则设置为不超过len(sorted_images)的2的幂
+        batch_size = min(batch_size, 2 ** (len(sorted_images).bit_length() - 1))
+
         dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=0)
 
         # Process batches and store results
@@ -115,7 +119,7 @@ class UnimernetModel(object):
                 mf_img = mf_img.to(dtype=self.model.dtype)
                 mf_img = mf_img.to(self.device)
                 with torch.no_grad():
-                    output = self.model.generate({"image": mf_img})
+                    output = self.model.generate({"image": mf_img}, batch_size=batch_size)
                 mfr_res.extend(output["fixed_str"])
 
                 # 更新进度条，每次增加batch_size，但要注意最后一个batch可能不足batch_size

mineru/model/mfr/unimernet/unimernet_hf/modeling_unimernet.py

@@ -468,7 +468,7 @@ class UnimernetModel(VisionEncoderDecoderModel):
         ).loss
         return {"loss": loss}
 
-    def generate(self, samples, do_sample: bool = False, temperature: float = 0.2, top_p: float = 0.95):
+    def generate(self, samples, do_sample: bool = False, temperature: float = 0.2, top_p: float = 0.95, batch_size=64):
         pixel_values = samples["image"]
         num_channels = pixel_values.shape[1]
         if num_channels == 1:
@@ -478,7 +478,13 @@ class UnimernetModel(VisionEncoderDecoderModel):
         if do_sample:
             kwargs["temperature"] = temperature
             kwargs["top_p"] = top_p
-        
+
+        if self.tokenizer.tokenizer.model_max_length > 1152:
+            if batch_size <= 32:
+                self.tokenizer.tokenizer.model_max_length = 1152  # 6g
+            else:
+                self.tokenizer.tokenizer.model_max_length = 1344  # 8g
+
         outputs = super().generate(
             pixel_values=pixel_values,
             max_new_tokens=self.tokenizer.tokenizer.model_max_length, # required

mineru/model/ocr/paddleocr2pytorch/pytorch_paddle.py

@@ -88,7 +88,7 @@ class PytorchPaddleOCR(TextSystem):
         kwargs['det_model_path'] = det_model_path
         kwargs['rec_model_path'] = rec_model_path
         kwargs['rec_char_dict_path'] = os.path.join(root_dir, 'pytorchocr', 'utils', 'resources', 'dict', dict_file)
-        # kwargs['rec_batch_num'] = 8
+        kwargs['rec_batch_num'] = 16
 
         kwargs['device'] = device
 

mineru/version.py

@@ -1 +1 @@
-__version__ = "2.1.0"
+__version__ = "2.1.1"

pyproject.toml

@@ -109,8 +109,10 @@ pipeline_old_linux = [
 ]
 
 [project.urls]
-Home = "https://mineru.net/"
-Repository = "https://github.com/opendatalab/MinerU"
+homepage = "https://mineru.net/"
+documentation = "https://opendatalab.github.io/MinerU/"
+repository = "https://github.com/opendatalab/MinerU"
+issues = "https://github.com/opendatalab/MinerU/issues"
 
 [project.scripts]
 mineru = "mineru.cli:client.main"