Merge pull request #285 from zhouzi180/master

Update for Scoliosis1K Dataset

datasets/Scoliosis1K/README.md

@@ -1,34 +1,85 @@
 # Tutorial for [Scoliosis1K](https://zhouzi180.github.io/Scoliosis1K)
+## Download the Scoliosis1K Dataset
 
-## Download the Scoliosis1K dataset
+You can download the dataset from the [official website](https://zhouzi180.github.io/Scoliosis1K).
-Download the dataset from the [link](https://zhouzi180.github.io/Scoliosis1K).
+The dataset is provided as four compressed files:
-decompress these two file by following command:
-```shell
-unzip -P password Scoliosis1K-pkl.zip   | xargs -n1 tar xzvf
-```
-password should be obtained by signing [agreement](https://zhouzi180.github.io/Scoliosis1K/static/resources/Scoliosis1KAgreement.pdf) and sending to email (12331257@mail.sustech.edu.cn)
 
-Then you will get Scoliosis1K formatted as:
+* `Scoliosis1K-sil-raw.zip`
-```
+* `Scoliosis1K-sil-pkl.zip`
-    DATASET_ROOT/
+* `Scoliosis1K-pose-raw.zip`
-        00000 (subject)/
+* `Scoliosis1K-pose-pkl.zip`
-            positive (category)/
+
-                    000-180 (view)/
+We recommend using the provided pickle (`.pkl`) files for convenience.
-                        000.pkl (contains all frames)
+Decompress them with the following commands:
-        ......
+
-```
+```bash
-## Train the dataset
+unzip -P <password> Scoliosis1K-sil-pkl.zip
-Modify the `dataset_root` in `configs/sconet/sconet_scoliosis1k.yaml`, and then run this command:
+unzip -P <password> Scoliosis1K-pose-pkl.zip
-```shell
-CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 opengait/main.py --cfgs configs/sconet/sconet_scoliosis1k.yaml --phase train
 ```
 
+> **Note**: The \<password\> can be obtained by signing the [release agreement](https://zhouzi180.github.io/Scoliosis1K/static/resources/Scoliosis1k_release_agreement.pdf) and sending it to **[12331257@mail.sustech.edu.cn](mailto:12331257@mail.sustech.edu.cn)**.
 
-## Process from RAW dataset
+### Dataset Structure
+
+After decompression, you will get the following structure:
 
-### Preprocess the dataset (Optional)
+```
-Download the raw dataset from the [official link](https://zhouzi180.github.io/Scoliosis1K). You will get two compressed files, i.e. `Scoliosis1K-raw.zip`, and `Scoliosis1K-pkl.zip`.
+├── Scoliosis1K-sil-pkl
-We recommend using our provided pickle files for convenience, or process raw dataset into pickle by this command:
+│   ├── 00000                     # Identity
-```shell
+│   │   ├── Positive              # Class
-python datasets/pretreatment.py --input_path Scoliosis1K_raw --output_path Scoliosis1K-pkl
+│   │   │   ├── 000_180           # View
+│   │   │   └── 000_180.pkl       # Estimated Silhouette (PP-HumanSeg v2)
+│
+├── Scoliosis1K-pose-pkl
+│   ├── 00000                     # Identity
+│   │   ├── Positive              # Class
+│   │   │   ├── 000_180           # View
+│   │   │   └── 000_180.pkl       # Estimated 2D Pose (ViTPose)
+```
+
+### Processing from RAW Dataset (optional)
+
+If you prefer, you can process the raw dataset into `.pkl` format.
+
+```bash
+# For silhouette raw data
+python datasets/pretreatment.py --input_path=<path_to_raw_silhouettes> -output_path=<output_path>
+
+# For pose raw data
+python datasets/pretreatment.py --input_path=<path_to_raw_pose> -output_path=<output_path> --pose --dataset=OUMVLP
+```
+---
+
+## Training and Testing
+
+Before training or testing, modify the `dataset_root` field in
+`configs/sconet/sconet_scoliosis1k.yaml`.
+
+Then run the following commands:
+
+```bash
+# Training
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+python -m torch.distributed.launch --nproc_per_node=4 \
+opengait/main.py --cfgs configs/sconet/sconet_scoliosis1k.yaml --phase train --log_to_file
+
+# Testing
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+python -m torch.distributed.launch --nproc_per_node=4 \
+opengait/main.py --cfgs configs/sconet/sconet_scoliosis1k.yaml --phase test --log_to_file
+```
+
+---
+
+## Pose-to-Heatmap Conversion
+
+*From our paper: **Pose as Clinical Prior: Learning Dual Representations for Scoliosis Screening (MICCAI 2025)***
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+python -m torch.distributed.launch --nproc_per_node=4 \
+datasets/pretreatment_heatmap.py \
+  --pose_data_path=<path_to_pose_pkl> \
+  --save_root=<output_path> \
+  --dataset_name=OUMVLP
 ```

opengait/evaluation/evaluator.py

@@ -5,7 +5,6 @@ from utils import get_msg_mgr, mkdir
 
 from .metric import mean_iou, cuda_dist, compute_ACC_mAP, evaluate_rank, evaluate_many
 from .re_rank import re_ranking
-from sklearn.metrics import confusion_matrix, accuracy_score
 
 def de_diag(acc, each_angle=False):
     # Exclude identical-view cases
@@ -417,46 +416,49 @@ def evaluate_CCPG(data, dataset, metric='euc'):
     return result_dict
 
 def evaluate_scoliosis(data, dataset, metric='euc'):
+
     msg_mgr = get_msg_mgr()
     
-    feature, label, class_id, view = data['embeddings'], data['labels'], data['types'], data['views']
+    from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score, confusion_matrix
 
-    label = np.array(label)
+    logits = np.array(data['embeddings'])
-    class_id = np.array(class_id)
+    labels = data['types']
     
-    # Update class_id with integer labels based on status
+    # Label mapping: negative->0, neutral->1, positive->2  
-    class_id_int = np.array([1 if status == 'positive' else 2 if status == 'neutral' else 0 for status in class_id])
+    label_map = {'negative': 0, 'neutral': 1, 'positive': 2}
-    print('class_id=', class_id_int)
+    true_ids = np.array([label_map[status] for status in labels])
     
-    features = np.array(feature)
+    pred_ids = np.argmax(logits.mean(-1), axis=-1)
-    c_id_int = np.argmax(features.mean(-1), axis=-1)
-    print('predicted_labels', c_id_int)
     
-    # Calculate sensitivity and specificity
+    # Calculate evaluation metrics
-    cm = confusion_matrix(class_id_int, c_id_int, labels=[0, 1, 2])
+    # Total Accuracy: proportion of correctly predicted samples among all samples
-    FP = cm.sum(axis=0) - np.diag(cm)
+    accuracy = accuracy_score(true_ids, pred_ids)
-    FN = cm.sum(axis=1) - np.diag(cm)
-    TP = np.diag(cm)
-    TN = cm.sum() - (FP + FN + TP)
     
-    # Sensitivity, hit rate, recall, or true positive rate
+    # Macro-average Precision: average of precision scores for each class
-    TPR = TP / (TP + FN)
+    precision = precision_score(true_ids, pred_ids, average='macro', zero_division=0)
-    # Specificity or true negative rate
-    TNR = TN / (TN + FP)
-    accuracy = accuracy_score(class_id_int, c_id_int)
     
-    result_dict = {}
+    # Macro-average Recall: average of recall scores for each class  
-    result_dict["scalar/test_accuracy/"] = accuracy
+    recall = recall_score(true_ids, pred_ids, average='macro', zero_division=0)
-    result_dict["scalar/test_sensitivity/"] = TPR
-    result_dict["scalar/test_specificity/"] = TNR
     
-    # Printing the sensitivity and specificity
+    # Macro-average F1: average of F1 scores for each class
-    for i, cls in enumerate(['Positive']):
+    f1 = f1_score(true_ids, pred_ids, average='macro', zero_division=0)
-        print(f"{cls} Sensitivity (Recall): {TPR[i] * 100:.2f}%")
-        print(f"{cls} Specificity: {TNR[i] * 100:.2f}%")
-    print(f"Accuracy: {accuracy * 100:.2f}%")
     
-    return result_dict
+    # Confusion matrix (for debugging)
+    # cm = confusion_matrix(true_ids, pred_ids, labels=[0, 1, 2])
+    # class_names = ['Negative', 'Neutral', 'Positive']
+    
+    # Print results
+    msg_mgr.log_info(f"Total Accuracy: {accuracy*100:.2f}%")
+    msg_mgr.log_info(f"Macro-avg Precision: {precision*100:.2f}%") 
+    msg_mgr.log_info(f"Macro-avg Recall: {recall*100:.2f}%")
+    msg_mgr.log_info(f"Macro-avg F1 Score: {f1*100:.2f}%")
+    
+    return {
+        "scalar/test_accuracy/": accuracy,
+        "scalar/test_precision/": precision, 
+        "scalar/test_recall/": recall,
+        "scalar/test_f1/": f1
+    }
 
 def evaluate_FreeGait(data, dataset, metric='euc'):
     msg_mgr = get_msg_mgr()

opengait/modeling/models/sconet.py

@@ -16,10 +16,10 @@ class ScoNet(BaseModel):
         self.HPP = HorizontalPoolingPyramid(bin_num=model_cfg['bin_num'])
 
     def forward(self, inputs):
-        ipts, labs, class_id, _, seqL = inputs
+        ipts, pids, labels, _, seqL = inputs
 
-        class_id_int = np.array([1 if status == 'positive' else 2 if status == 'neutral' else 0 for status in class_id])
+        # Label mapping: negative->0, neutral->1, positive->2 
-        class_id = torch.tensor(class_id_int).cuda()
+        label_ids = np.array([{'negative': 0, 'neutral': 1, 'positive': 2}[status] for status in labels])
 
         sils = ipts[0]
         if len(sils.size()) == 4:
@@ -40,8 +40,8 @@ class ScoNet(BaseModel):
         embed = embed_1
         retval = {
             'training_feat': {
-                'triplet': {'embeddings': embed, 'labels': labs},
+                'triplet': {'embeddings': embed, 'labels': pids},
-                'softmax': {'logits': logits, 'labels': class_id},
+                'softmax': {'logits': logits, 'labels': label_ids},
             },
             'visual_summary': {
                 'image/sils': rearrange(sils,'n c s h w -> (n s) c h w')