Merge pull request #2 from wang-tf:develop

Develop

.github/workflows/python-publish.yml

@@ -0,0 +1,28 @@
+name: deploy
+
+on: push
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build-n-publish:
+    runs-on: ubuntu-latest
+    if: startsWith(github.event.ref, 'refs/tags')
+    steps:
+      - uses: actions/checkout@v2
+      - name: Set up Python 3.7
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.7
+      - name: Install torch
+        run: pip install torch
+      - name: Install wheel
+        run: pip install wheel
+      - name: Build Deep Vital
+        run: python setup.py sdist bdist_wheel
+      - name: Publish distribution to PyPI
+        run: |
+          pip install twine
+          twine upload dist/* -u __token__ -p ${{ secrets.pypi_password }}

README.md

@@ -1,4 +1,8 @@
-# deep_vital
+# Deep Vital
 
 Pytorch version of [non-invasive-bp-estimation-using-deep-learning](https://github.com/Fabian-Sc85/non-invasive-bp-estimation-using-deep-learning)
 
+pretrained model can be found from [zendo](https://zenodo.org/record/7948098)
+
+## Models
+- [ResNet](configs/resnet/README.md)

configs/resnet/README.md

@@ -0,0 +1,23 @@
+# ResNet Blood Pressure Model
+
+> [Assessment of Non-Invasive Blood Pressure Prediction from PPG and rPPG Signals Using Deep Learning](https://readpaper.com/paper/3198128029)
+
+## Abstract
+
+## Results and Models
+| Model | MAE-SBP | MAE-DBP |
+| :---: | :-----: | :-----: |
+| ResNet501D | - | - |
+
+
+## Citation
+
+```latex
+@inproceedings{schrumpf2021assessment,
+  title={Assessment of deep learning based blood pressure prediction from PPG and rPPG signals},
+  author={Schrumpf, Fabian and Frenzel, Patrick and Aust, Christoph and Osterhoff, Georg and Fuchs, Mirco},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={3820--3830},
+  year={2021}
+}
+```

configs/resnet50_1xb16_bp.py → configs/resnet/resnet50_1xb16_bp_rppg.py

@@ -1,14 +1,14 @@
-_base_ = ['_base_/default_runtime.py']
+_base_ = ['../_base_/default_runtime.py']
 
 model = dict(type='BPResNet1D',
-             data_preprocessor=dict(type='RppgDataPreprocessor'),
+             data_preprocessor=dict(type='DataPreprocessor'),
              backbone=dict(
                  type='ResNet1D',
                  depth=50,
                  frozen_stages=4,
                  init_cfg=dict(
                      type='Pretrained',
-                     checkpoint='data/resnet_ppg_nonmixed_backbone_v2.pb')),
+                     checkpoint='data/resnet_ppg_nonmixed_backbone.pth')),
              neck=dict(type='AveragePooling'),
              head=dict(type='BPDenseHead', loss=dict(type='MSELoss')))
 
@@ -20,11 +20,11 @@
     dict(type='PackInputs', input_key='rppg'),
 ]
 train_dataloader = dict(
-    batch_size=16,
+    batch_size=32,
     num_workers=2,
     dataset=dict(type=dataset_type,
                  ann_file='data/rPPG-BP-UKL_rppg_7s.h5',
-                 used_subjects=[7.,  8., 10., 13., 14., 19., 21., 23., 24., 27., 33., 35., 36., 44., 46., 48.],
+                 used_idx_file='data/train.txt',
                  data_prefix='',
                  test_mode=False,
                  pipeline=train_pipeline),
@@ -35,14 +35,24 @@
     num_workers=2,
     dataset=dict(type=dataset_type,
                  ann_file='data/rPPG-BP-UKL_rppg_7s.h5',
-                 used_subjects=[6],
+                 used_idx_file='data/val.txt',
                  data_prefix='',
                  test_mode=True,
                  pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )
 val_evaluator = dict(type='MAE', gt_key='gt_label', pred_key='pred_label')
-test_dataloader = val_dataloader
+test_dataloader = dict(
+    batch_size=16,
+    num_workers=2,
+    dataset=dict(type=dataset_type,
+                 ann_file='data/rPPG-BP-UKL_rppg_7s.h5',
+                 used_idx_file='data/test.txt',
+                 data_prefix='',
+                 test_mode=True,
+                 pipeline=test_pipeline),
+    sampler=dict(type='DefaultSampler', shuffle=False),
+)
 test_evaluator = val_evaluator
 
 # optimizer
@@ -65,3 +75,8 @@
         dict(type='TensorboardVisBackend')
     ],
 )
+
+default_hooks = dict(
+    checkpoint=dict(type='CheckpointHook', interval=1, by_epoch=True, save_best='loss', rule='less'),
+)
+custom_hooks = [dict(type='EarlyStoppingHook', monitor='loss', rule='less', min_delta=0.01, strict=False, check_finite=True, patience=5)]

configs/resnet/resnet50_4xb128_bp_ppg.py

@@ -0,0 +1,87 @@
+_base_ = ['../_base_/default_runtime.py']
+
+model = dict(type='BPResNet1D',
+             data_preprocessor=dict(type='DataPreprocessor'),
+             backbone=dict(
+                 type='ResNet1D',
+                 depth=50,
+             ),
+             neck=dict(type='AveragePooling'),
+             head=dict(type='BPDenseHead', loss=dict(type='MSELoss')))
+
+dataset_type = 'PpgData'
+train_pipeline = [
+    dict(type='PackInputs', input_key='ppg'),
+]
+test_pipeline = [
+    dict(type='PackInputs', input_key='ppg'),
+]
+train_dataloader = dict(
+    batch_size=128,
+    num_workers=2,
+    dataset=dict(type=dataset_type,
+                 ann_file='data/mimic-iii_data/train.h5',
+                 data_prefix='',
+                 test_mode=False,
+                 pipeline=train_pipeline),
+    sampler=dict(type='DefaultSampler', shuffle=True),
+)
+val_dataloader = dict(
+    batch_size=64,
+    num_workers=2,
+    dataset=dict(type=dataset_type,
+                 ann_file='data/mimic-iii_data/val.h5',
+                 data_prefix='',
+                 test_mode=True,
+                 pipeline=test_pipeline),
+    sampler=dict(type='DefaultSampler', shuffle=False),
+)
+val_evaluator = [
+    dict(type='MAE', gt_key='gt_label',
+         pred_key='pred_label', compute_loss=True),
+    # dict(type='BlandAltmanPlot')
+]
+test_dataloader = dict(
+    batch_size=64,
+    num_workers=2,
+    dataset=dict(type=dataset_type,
+                 ann_file='data/mimic-iii_data/test.h5',
+                 data_prefix='',
+                 test_mode=True,
+                 pipeline=test_pipeline),
+    sampler=dict(type='DefaultSampler', shuffle=False),
+)
+test_evaluator = [
+    dict(type='MAE', gt_key='gt_label',
+         pred_key='pred_label')
+]
+
+# optimizer
+optim_wrapper = dict(optimizer=dict(type='Adam',
+                                    lr=0.001,
+                                    betas=(0.9, 0.999),
+                                    eps=1e-08,
+                                    weight_decay=0,
+                                    amsgrad=False))
+
+# train, val, test setting
+train_cfg = dict(by_epoch=True, max_epochs=200, val_interval=1)
+val_cfg = dict()
+test_cfg = dict()
+
+visualizer = dict(
+    type='UniversalVisualizer',
+    vis_backends=[
+        dict(type='LocalVisBackend'),
+        dict(type='TensorboardVisBackend')
+    ],
+)
+
+default_hooks = dict(
+    checkpoint=dict(type='CheckpointHook', interval=1,
+                    by_epoch=True, save_best='MAE/pred_loss', rule='less'),
+)
+custom_hooks = [
+    dict(type='EarlyStoppingHook', monitor='MAE/pred_loss', rule='less',
+         min_delta=0.01, strict=False, check_finite=True, patience=5)
+]

deep_vital/datasets/__init__.py

@@ -1,2 +1,3 @@
-from .ppg_signal_dataset import RppgData
+from .ppg_signal_dataset import PpgData
+from .rppg_signal_dataset import RppgData
 from .transforms import *

deep_vital/datasets/ppg_signal_dataset.py

@@ -17,10 +17,11 @@ def expanduser(path):
 
 
 @DATASETS.register_module()
-class RppgData(BaseDataset):
+class PpgData(BaseDataset):
+    _signal_name = 'ppg'
     def __init__(self,
                  ann_file,
-                 used_subjects=(),
+                 used_idx_file=None,
                  metainfo=None,
                  data_root='',
                  data_prefix='',
@@ -32,9 +33,9 @@ def __init__(self,
                  lazy_init=False,
                  max_refetch=1000):
         self.label = None
-        self.rppg = None
+        self.signal = None
         self.subject_idx = None
-        self.used_subjects=used_subjects
+        self.used_idx_file=used_idx_file
         self.num = None
 
         if isinstance(data_prefix, str):
@@ -65,24 +66,24 @@ def load_data_list(self):
         assert os.path.exists(self.ann_file), self.ann_file
 
         data = h5py.File(self.ann_file, 'r')
-        self.label = np.array(data.get('/label')).T
-        self.rppg = np.array(data.get('/rppg')).T
-        self.subject_idx = np.array(data.get('/subject_idx'), dtype=int)[0, :]
+        self.label = np.array(data.get('/label'))
+        self.signal = np.array(data.get(f'/{self._signal_name}'))
+        self.subject_idx = np.array(data.get('/subject_idx'), dtype=int)
         subjects_list = np.unique(self.subject_idx)
 
-        if self.used_subjects:
-            idx_used = np.where(np.isin(self.subject_idx, self.used_subjects))[-1]
+        if self.used_idx_file and os.path.exists(self.used_idx_file):
+            idx_used = np.loadtxt(self.used_idx_file, dtype=np.int64, delimiter=',')
             self.label = self.label[idx_used]
-            self.rppg = self.rppg[idx_used]
+            self.signal = self.signal[idx_used]
             self.subject_idx = self.subject_idx[idx_used]
-        
+
         self.num = self.subject_idx.shape[0]
-        
+
         data_list = []
-        for _label, _rppg, _subject_idx in zip(self.label, self.rppg, self.subject_idx):
+        for _label, _signal, _subject_idx in zip(self.label, self.signal, self.subject_idx):
             # sbp_label = _label[0]
             # dbp_label = _label[1]
             # _label = _label.reshape((1, 2, 1))
-            info = {'gt_label': _label, 'rppg': _rppg, 'subject_idx': _subject_idx}
+            info = {'gt_label': _label, f'{self._signal_name}': _signal, 'subject_idx': _subject_idx}
             data_list.append(info)
         return data_list

deep_vital/datasets/rppg_signal_dataset.py

@@ -0,0 +1,8 @@
+
+from deep_vital.registry import DATASETS
+from .ppg_signal_dataset import PpgData
+
+
+@DATASETS.register_module()
+class RppgData(PpgData):
+    _signal_name = 'rppg'

deep_vital/evaluation/metrics/__init__.py

@@ -1 +1,2 @@
-from .mae import MAE
+from .mae import MAE
+from .bland_altman_plot import BlandAltmanPlot

deep_vital/evaluation/metrics/bland_altman_plot.py

@@ -0,0 +1,74 @@
+from typing import Optional
+import torch
+import numpy as np
+import matplotlib.pyplot as plt
+from matplotlib.backends.backend_agg import FigureCanvasAgg
+import PIL.Image as Image
+from mmengine.evaluator import BaseMetric
+from mmengine.visualization import Visualizer
+from mmengine import MessageHub
+from deep_vital.registry import METRICS
+
+message_hub = MessageHub.get_current_instance()
+
+
+def bland_altman_plot(data1, data2, *args, **kwargs):
+    data1 = np.asarray(data1.numpy())
+    data2 = np.asarray(data2.numpy())
+    mean = np.mean([data1, data2], axis=0)
+    diff = data1 - data2                   # Difference between data1 and data2
+    md = np.mean(diff)                   # Mean of the difference
+    sd = np.std(diff, axis=0)            # Standard deviation of the difference
+
+    plt.cla()
+    plt.scatter(mean, diff, *args, **kwargs)
+    plt.axhline(md,           color='gray', linestyle='-')
+    plt.axhline(md + 1.96*sd, color='gray', linestyle='--')
+    plt.axhline(md - 1.96*sd, color='gray', linestyle='--')
+    canvas = FigureCanvasAgg(plt.gcf())
+    canvas.draw()
+    w, h = canvas.get_width_height()
+    buf = np.fromstring(canvas.tostring_argb(), dtype=np.uint8)
+    buf.shape = (w, h, 4)
+    buf = np.roll(buf, 3, axis=2)
+    image = Image.frombytes('RGBA', (w, h), buf.tostring())
+    image = np.asarray(image)
+    rgb_image = image[:, :, :3]
+    plt.close()
+    return rgb_image
+
+
+@METRICS.register_module()
+class BlandAltmanPlot(BaseMetric):
+    metric = 'BlandAltmanPlot'
+    default_prefix = 'BlandAltmanPlot'
+
+    def __init__(self,
+                 gt_key='gt_label',
+                 pred_key='pred_label',
+                 collect_device: str = 'cpu',
+                 prefix: Optional[str] = None):
+        super().__init__(collect_device=collect_device, prefix=prefix)
+        self.gt_key = gt_key
+        self.pred_key = pred_key
+        self.visualizer = Visualizer.get_current_instance()
+
+    def process(self, data_batch, data_samples):
+        for data_sample in data_samples:
+            result = {
+                'pred_label': data_sample[self.pred_key],
+                'gt_label': data_sample[self.gt_key],
+            }
+            self.results.append(result)
+
+    def compute_metrics(self, results):
+        metrics = {}
+        target = torch.stack([res['gt_label'] for res in results])
+        pred = torch.stack([res['pred_label'] for res in results])
+
+        sbp_BAP = bland_altman_plot(pred[:, 0], target[:, 0])
+        dbp_BAP = bland_altman_plot(pred[:, 1], target[:, 1])
+        current_step = message_hub.get_info('epoch')
+        self.visualizer.add_image('sbp_BAP', sbp_BAP, current_step)
+        self.visualizer.add_image('dbp_BAP', dbp_BAP, current_step)
+        return metrics