Replace colour model with binary lesion model v0.2.0

Final 51-epoch model trained on all 1,465 images; outputs background/lesion only.

README.md

@@ -6,7 +6,7 @@ library_name: cervicalseg
 tags:
   - medical-imaging
   - cervical
-  - semantic-segmentation
+  - binary-segmentation
   - dinov3
   - dpt
   - wtconv
@@ -14,80 +14,76 @@ tags:
 
 # CervicalSeg
 
-CervicalSeg is a four-class semantic segmentation model for cervical colposcopy images. It combines
-a DINOv3 ViT-S/16 backbone, a DPT decoder, and WTConv in the final two high-resolution decoder
-fusion stages.
+CervicalSeg is a binary lesion segmentation model for cervical colposcopy images. It combines a
+DINOv3 ViT-S/16 backbone, a DPT decoder, and WTConv in the final two high-resolution decoder fusion
+stages. It predicts the lesion region only and does not classify red, blue, or green annotation
+categories.
 
-宫颈阴道镜图像四分类语义分割模型，采用 DINOv3 ViT-S/16、DPT 解码器，并在最后两个高分辨率
-融合阶段使用 WTConv。
+CervicalSeg 是宫颈阴道镜图像二分类病灶分割模型，采用 DINOv3 ViT-S/16、DPT 解码器，并在最后
+两个高分辨率融合阶段使用 WTConv。模型只预测病灶区域，不再区分红、蓝、绿标注类别。
 
 ## Classes / 类别
 
-| Index | Class | Colour |
+| Index | Class | Meaning |
 |---:|---|---|
-| 0 | background | black |
-| 1 | blue | blue |
-| 2 | green | green |
-| 3 | red | red |
+| 0 | background | non-lesion region / 非病灶区域 |
+| 1 | lesion | lesion region / 病灶区域 |
 
 ## Usage / 使用方法
 
 ```bash
-pip install cervicalseg
+pip install "cervicalseg>=0.2.0"
 ```
 
-Use `cervicalseg>=0.1.1` for compatibility with both Transformers 4.56+ and 5.x DINOv3 module
-layouts.
-
-请使用 `cervicalseg>=0.1.1`，以兼容 Transformers 4.56+ 与 5.x 的两种 DINOv3 模块命名。
-
 ```python
 from cervicalseg import CervicalSeg
 
 segmenter = CervicalSeg(device="auto")
 result = segmenter.predict("image.jpg")
-result.save_mask("mask.png")
-result.save_color_mask("color_mask.png")
+
+# result.mask contains only 0=background and 1=lesion.
+result.save_mask("mask.png")       # binary PNG containing 0 and 255
 result.save_overlay("overlay.jpg")
 ```
 
 The first call downloads `model.safetensors`; later calls reuse the Hugging Face cache. Output masks
-are restored to the original image size.
+are restored to the original image size. The single overlay colour is for visualization only and is
+not a lesion subtype.
 
 首次调用会下载 `model.safetensors`，后续调用复用 Hugging Face 缓存。输出 mask 会恢复到原图尺寸。
+叠加图中的单一颜色仅用于显示病灶范围，不代表病灶类别。
 
-## Evaluation / 评估结果
-
-The data were split at the patient level to prevent patient leakage between train, validation, and
-test sets.
+## Training and evaluation / 训练与评估
 
-数据按患者划分，避免同一患者同时出现在训练、验证或测试集合中。
+The public v0.2.0 checkpoint was trained for 51 epochs on all 1,465 available images: 1,028 from the
+original training split, 221 from validation, and 216 from test. All non-background annotations were
+merged into one lesion label. Because all samples were used to fit the final model, no held-out test
+metric applies to this public checkpoint.
 
-| Split | Samples | Foreground mIoU | Foreground Dice | Pixel accuracy |
-|---|---:|---:|---:|---:|
-| Validation | 221 | 0.4075 | 0.5704 | 0.9121 |
-| Test | 216 | 0.3633 | 0.5296 | 0.8891 |
+公开的 v0.2.0 权重使用全部 1,465 张图像训练 51 个 epoch，其中原 train/val/test 分别为
+1,028/221/216 张。所有非背景标注合并为一个病灶标签。由于最终模型使用了全部样本训练，因此该
+公开权重没有独立测试集指标。
 
-Test-set per-class metrics:
+For model-development reference only, the separate patient-level validation experiment at epoch 51
+obtained lesion IoU 0.5847 and lesion Dice 0.7379. These are development results, not an independent
+evaluation of the all-data public checkpoint.
 
-| Class | Precision | Recall | Dice | IoU |
-|---|---:|---:|---:|---:|
-| blue | 0.5651 | 0.5697 | 0.5674 | 0.3961 |
-| green | 0.3784 | 0.4950 | 0.4289 | 0.2730 |
-| red | 0.6850 | 0.5219 | 0.5924 | 0.4209 |
+仅供模型开发参考：按患者划分的独立验证实验在第 51 个 epoch 得到病灶 IoU 0.5847、病灶 Dice
+0.7379；这些结果不是对全量训练公开权重的独立测试。
 
 ## Intended use and limitations / 用途与限制
 
 - Research use only.
 - Not a medical device and not for clinical diagnosis or treatment decisions.
 - Predictions require review by qualified professionals.
+- The model detects lesion extent but does not predict lesion grade, pathology, or colour category.
 - Performance may not generalize to devices, institutions, populations, or acquisition conditions
   not represented in the training data.
-- Green-region segmentation is the weakest class in the reported test results.
 
 - 仅用于研究。
 - 不是医疗器械，不得直接用于临床诊断或治疗决策。
 - 预测结果需要由具备资质的专业人员复核。
+- 模型仅检测病灶范围，不预测病灶分级、病理结果或颜色类别。
 - 对训练数据未覆盖的设备、机构、人群和采集条件，模型性能可能下降。
 
 ## License / 许可证

config.json

@@ -1,11 +1,9 @@
 {
-  "format_version": 1,
+  "format_version": 2,
   "model": {
     "class_names": [
       "background",
-      "blue",
-      "green",
-      "red"
+      "lesion"
     ],
     "image_size": [
       576,
@@ -75,11 +73,23 @@
   },
   "checkpoint": {
     "filename": "model.safetensors",
-    "size_bytes": 96876480,
-    "sha256": "2fdec883d6a94b7a3478a750f6943bb41b60abcb8316649bd8a050f9bd3e5f4d",
-    "training_epoch": 66,
-    "best_validation_miou_foreground": 0.40750709018075176,
-    "test_miou_foreground": 0.3633164423801378,
-    "test_mean_dice_foreground": 0.5295766998594339
+    "size_bytes": 96875968,
+    "sha256": "2322f48bf36f0fdf77f090c5621c8d31aa6892b51000d40e42f7f43b23c29b98",
+    "training_epoch": 51,
+    "checkpoint_selection": "final_epoch",
+    "training_splits": [
+      "train",
+      "val",
+      "test"
+    ],
+    "training_sample_counts": {
+      "train": 1028,
+      "val": 221,
+      "test": 216
+    },
+    "training_samples_total": 1465,
+    "final_training_loss": 0.08957859227885158,
+    "development_validation_foreground_iou_epoch_51": 0.5847087461680525,
+    "development_validation_foreground_dice_epoch_51": 0.7379384351628312
   }
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fdec883d6a94b7a3478a750f6943bb41b60abcb8316649bd8a050f9bd3e5f4d
-size 96876480
+oid sha256:2322f48bf36f0fdf77f090c5621c8d31aa6892b51000d40e42f7f43b23c29b98
+size 96875968