Pytorch dev

caiman/components_evaluation.py

@@ -6,7 +6,7 @@
 import numpy as np
 import os
 import peakutils
-import tensorflow as tf
+import torch 
 import scipy
 from scipy.sparse import csc_matrix
 from scipy.stats import norm
@@ -273,42 +273,37 @@ def evaluate_components_CNN(A,
     if not isGPU and 'CAIMAN_ALLOW_GPU' not in os.environ:
         print("GPU run not requested, disabling use of GPUs")
         os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
-    try:
-        os.environ["KERAS_BACKEND"] = "tensorflow"
-        from tensorflow.keras.models import model_from_json
-        use_keras = True
-        logger.info('Using Keras')
+    try: 
+        os.environ["KERAS_BACKEND"] = "torch"
+        from keras.models import model_load 
+        use_keras = True 
+        logging.info('Using Keras')
     except (ModuleNotFoundError):
-        use_keras = False
-        logger.info('Using Tensorflow')
+        use_keras = False 
+        logging.info('Using Torch')
 
     if loaded_model is None:
         if use_keras:
-            if os.path.isfile(os.path.join(caiman_datadir(), model_name + ".json")):
-                model_file = os.path.join(caiman_datadir(), model_name + ".json")
-                model_weights = os.path.join(caiman_datadir(), model_name + ".h5")
-            elif os.path.isfile(model_name + ".json"):
-                model_file = model_name + ".json"
-                model_weights = model_name + ".h5"
+            if os.path.isfile(os.path.join(caiman_datadir(), model_name + ".keras")):
+                model_file = os.path.join(caiman_datadir(), model_name + ".keras")
+            elif os.path.isfile(model_name + ".keras"):
+                model_file = model_name + ".keras"
             else:
                 raise FileNotFoundError(f"File for requested model {model_name} not found")
-            with open(model_file, 'r') as json_file:
-                print(f"USING MODEL (keras API): {model_file}")
-                loaded_model_json = json_file.read()
-
-            loaded_model = model_from_json(loaded_model_json)
-            loaded_model.load_weights(model_name + '.h5')
+
+            print(f"USING MODEL (keras API): {model_file}")
+            loaded_model = model_load(model_file)
         else:
-            if os.path.isfile(os.path.join(caiman_datadir(), model_name + ".h5.pb")):
-                model_file = os.path.join(caiman_datadir(), model_name + ".h5.pb")
-            elif os.path.isfile(model_name + ".h5.pb"):
-                model_file = model_name + ".h5.pb"
+            if os.path.isfile(os.path.join(caiman_datadir(), model_name + ".pt")):
+                model_file = os.path.join(caiman_datadir(), model_name + ".pt")
+            elif os.path.isfile(model_name + ".pt"):
+                model_file = model_name + ".pt"
             else:
                 raise FileNotFoundError(f"File for requested model {model_name} not found")
             print(f"USING MODEL (tensorflow API): {model_file}")
-            loaded_model = caiman.utils.utils.load_graph(model_file)
+            loaded_model = torch.load(model_file)
 
-        logger.debug("Loaded model from disk")
+        logging.debug("Loaded model from disk")
 
     half_crop = np.minimum(gSig[0] * 4 + 1, patch_size), np.minimum(gSig[1] * 4 + 1, patch_size)
     dims = np.array(dims)
@@ -323,11 +318,11 @@ def evaluate_components_CNN(A,
     if use_keras:
         predictions = loaded_model.predict(final_crops[:, :, :, np.newaxis], batch_size=32, verbose=1)
     else:
-        tf_in = loaded_model.get_tensor_by_name('prefix/conv2d_20_input:0')
-        tf_out = loaded_model.get_tensor_by_name('prefix/output_node0:0')
-        with tf.Session(graph=loaded_model) as sess:
-            predictions = sess.run(tf_out, feed_dict={tf_in: final_crops[:, :, :, np.newaxis]})
-            sess.close()
+        final_crops = torch.tensor(final_crops, dtype=torch.float32)
+        final_crops = torch.reshape(final_crops, (-1, final_crops.shape[-1], 
+                        final_crops.shape[1], final_crops.shape[2])) 
+        with torch.no_grad():
+            prediction = loaded_model(final_crops[:, np.newaxis, :, :])
 
     return predictions, final_crops
 

caiman/source_extraction/cnmf/online_cnmf.py

@@ -13,6 +13,9 @@
 imaging data in real time. In Advances in Neural Information Processing Systems 
 (pp. 2381-2391).
 @url http://papers.nips.cc/paper/6832-onacid-online-analysis-of-calcium-imaging-data-in-real-time
+
+Implemented in PyTorch
+Date: July 18, 2024
 """
 
 import cv2
@@ -26,7 +29,7 @@
 from scipy.stats import norm
 from sklearn.decomposition import NMF
 from sklearn.preprocessing import normalize
-import tensorflow as tf
+import torch 
 from time import time
 
 import caiman
@@ -322,32 +325,28 @@ def _prepare_object(self, Yr, T, new_dims=None, idx_components=None):
             self.params.set('online', {'sniper_mode': False})
             self.tf_in = None
             self.tf_out = None
+            # self.use_torch = None 
         else:
-            try:
-                from tensorflow.keras.models import model_from_json
-                logger.info('Using Keras')
+            try: 
+                from keras.models import load_model 
+                logging.info('Using Keras')
                 use_keras = True
             except(ModuleNotFoundError):
-                use_keras = False
-                logger.info('Using Tensorflow')
+                use_keras = False 
+                logging.info('Using Torch')
+
+            path = self.params.get('online', 'path_to_model').split(".")[:-1]
             if use_keras:
-                path = self.params.get('online', 'path_to_model').split(".")[:-1]
-                json_path = ".".join(path + ["json"])
-                model_path = ".".join(path + ["h5"])
-                json_file = open(json_path, 'r')
-                loaded_model_json = json_file.read()
-                json_file.close()
-                loaded_model = model_from_json(loaded_model_json)
-                loaded_model.load_weights(model_path)
-                self.tf_in = None
-                self.tf_out = None
-            else:
-                path = self.params.get('online', 'path_to_model').split(".")[:-1]
-                model_path = '.'.join(path + ['h5', 'pb'])
+                # uses online model -> be careful 
+                model_path = ".".join(path + ["keras"])
+                loaded_model = model_load(model_path)
+                # self.use_torch = False 
+            else: 
+                model_path = '.'.join(path + ['pt'])
                 loaded_model = load_graph(model_path)
-                self.tf_in = loaded_model.get_tensor_by_name('prefix/conv2d_1_input:0')
-                self.tf_out = loaded_model.get_tensor_by_name('prefix/output_node0:0')
-                loaded_model = tf.Session(graph=loaded_model)
+                loaded_model = torch.load(model_file)
+                # self.use_torch = True 
+
         self.loaded_model = loaded_model
 
         if self.is1p:
@@ -549,6 +548,7 @@ def fit_next(self, t, frame_in, num_iters_hals=3):
                 use_peak_max=self.params.get('online', 'use_peak_max'),
                 mean_buff=self.estimates.mean_buff,
                 tf_in=self.tf_in, tf_out=self.tf_out,
+                # use_torch=self.use_torch,
                 ssub_B=ssub_B, W=self.estimates.W if self.is1p else None,
                 b0=self.estimates.b0 if self.is1p else None,
                 corr_img=self.estimates.corr_img if use_corr else None,
@@ -2004,6 +2004,7 @@ def get_candidate_components(sv, dims, Yres_buf, min_num_trial=3, gSig=(5, 5),
                              thresh_CNN_noisy=0.5, use_peak_max=False,
                              thresh_std_peak_resid = 1, mean_buff=None,
                              tf_in=None, tf_out=None):
+                             # use_torch=None):
     """
     Extract new candidate components from the residual buffer and test them
     using space correlation or the CNN classifier. The function runs the CNN
@@ -2084,12 +2085,19 @@ def get_candidate_components(sv, dims, Yres_buf, min_num_trial=3, gSig=(5, 5),
         Ain2 /= np.std(Ain2,axis=1)[:,None]
         Ain2 = np.reshape(Ain2,(-1,) + tuple(np.diff(ijSig_cnn).squeeze()),order= 'F')
         Ain2 = np.stack([cv2.resize(ain,(patch_size ,patch_size)) for ain in Ain2])
-        if tf_in is None:
+        if use_torch is None:
             predictions = loaded_model.predict(Ain2[:,:,:,np.newaxis], batch_size=min_num_trial, verbose=0)
+            keep_cnn = list(np.where(predictions[:, 0] > thresh_CNN_noisy)[0])
         else:
-            predictions = loaded_model.run(tf_out, feed_dict={tf_in: Ain2[:, :, :, np.newaxis]})
-        keep_cnn = list(np.where(predictions[:, 0] > thresh_CNN_noisy)[0])
-        cnn_pos = Ain2[keep_cnn]
+            final_crops = torch.tensor(Ain2, dtype=torch.float32)
+            final_crops = torch.reshape(Ain2, (-1, Ain2.shape[-1], 
+                        Ain2.shape[1], Ain2.shape[2])) 
+            with torch.no_grad():
+                prediction = loaded_model(Ain2[:, np.newaxis, :, :])     
+            keep_cnn = list(torch.where(predictions[:, 0] > thresh_CNN_noisy)[0])
+
+        cnn_pos = Ain2[keep_cnn] #Make sure this works 
+        # tensor.numpy() also works 
     else:
         keep_cnn = []  # list(range(len(Ain_cnn)))
 
@@ -2139,6 +2147,7 @@ def update_num_components(t, sv, Ab, Cf, Yres_buf, Y_buf, rho_buf,
                           crosscorr=None, col_ind=None, row_ind=None, corr_img_mode=None,
                           max_img=None, downscale_matrix=None, upscale_matrix=None,
                           tf_in=None, tf_out=None):
+                          # torch_in=None, torch_out=None):
     """
     Checks for new components in the residual buffer and incorporates them if they pass the acceptance tests
     """
@@ -2169,6 +2178,7 @@ def update_num_components(t, sv, Ab, Cf, Yres_buf, Y_buf, rho_buf,
         loaded_model=loaded_model, thresh_CNN_noisy=thresh_CNN_noisy,
         use_peak_max=use_peak_max, test_both=test_both, mean_buff=mean_buff,
         tf_in=tf_in, tf_out=tf_out)
+        #torch_in=torch_in, torch_out=torch_out)
 
     ind_new_all = ijsig_all
 

caiman/tests/test_pytorch.py

@@ -0,0 +1,49 @@
+#!/usr/bin/env python
+
+import numpy as np
+import os
+import keras 
+
+from caiman.paths import caiman_datadir
+from caiman.utils.utils import load_graph
+
+try:
+    os.environ["KERAS_BACKEND"] = "torch"
+    from keras.models import load_model 
+    use_keras = True
+except(ModuleNotFoundError):
+    import torch 
+    use_keras = False
+
+def test_torch():
+    os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
+
+    try:
+        model_name = os.path.join(caiman_datadir(), 'model', 'cnn_model')
+        if use_keras:
+            model_file = model_name + ".keras"
+            print('USING MODEL:' + model_file)
+
+            loaded_model = load_model(model_file)
+            loaded_model.compile('sgd', 'mse')
+        elif use_keras == True: 
+            model_file = model_name + ".pth"
+            loaded_model = torch.load(model_file)
+    except:
+        raise Exception(f'NN model could not be loaded. use_keras = {use_keras}')
+
+    A = np.random.randn(10, 50, 50, 1)
+    try:
+        if use_keras == False: 
+            predictions = loaded_model.predict(A, batch_size=32)
+        elif use_keras == True:
+            A = torch.tensor(A, dtype=torch.float32)
+            A = torch.reshape(A, (-1, A.shape[-1], A.shape[1], A.shape[2])) 
+            with torch.no_grad():
+                predictions = loaded_model(A)  
+        pass 
+    except:
+        raise Exception('NN model could not be deployed. use_keras = ' + str(use_keras))
+
+if __name__ == "__main__":
+    test_torch()

caiman/train/__init__.py

@@ -0,0 +1,7 @@
+#!/usr/bin/env python
+import pkg_resources
+
+from caiman.train.train_cnn_model_keras import cnn_model_keras, save_model_keras, load_model_keras
+from caiman.train.train_cnn_model_pytorch import cnn_model_pytorch, train_test_split, train, validate, get_batch_accuracy, save_model_pytorch, load_model_pytorch
+
+__version__ = pkg_resources.get_distribution('caiman').version