Merge pull request #53 from RichieHakim/dev

Dev

batch_run/dispatcher.py

@@ -35,7 +35,7 @@
         "overwrite_config": True,
         "update_project_paths": True,
         "initialize_visualization": False,
-        "use_GPU": False,
+        "use_GPU": True,
         "random_seed": 0,
         "verbose": 2,
     },
@@ -84,16 +84,16 @@
         },
     },
     "PointTracker": {
-        "contiguous": False,
+        "contiguous": True,
         "params_optical_flow": {
             "method": "lucas_kanade",
-            "mesh_rigidity": 0.025,
-            "mesh_n_neighbors": 8,
+            "mesh_rigidity": 0.045,
+            "mesh_n_neighbors": 12,
             "relaxation": 0.0015,
             "kwargs_method": {
                 "winSize": [
-                    20,
-                    20,
+                    25,
+                    25,
                 ],
                 "maxLevel": 2,
                 "criteria": [
@@ -110,19 +110,19 @@
         },
         "params_outlier_handling": {
             "threshold_displacement": 150,
-            "framesHalted_before": 10,
-            "framesHalted_after": 10,
+            "framesHalted_before": 40,
+            "framesHalted_after": 40,
         },
         "verbose": 2,
     },
     "VQT_Analyzer": {
         "params_VQT": {
-            "Q_lowF": 4,
-            "Q_highF": 10,
+            "Q_lowF": 3,
+            "Q_highF": 5,
             "F_min": 1.0,
             "F_max": 60,
             "n_freq_bins": 36,
-            "win_size": 501,
+            "win_size": 701,
             "symmetry": 'left',
             "taper_asymmetric": True,
             "plot_pref": False,
@@ -163,7 +163,7 @@
         "fit": {
             "method": "CP_NN_HALS",
             "params_method": {
-                "rank": 12,
+                "rank": 10,
                 "n_iter_max": 200,
                 "init": "random",
                 "svd": "truncated_svd",
@@ -241,12 +241,12 @@
 #SBATCH --job-name={name_slurm}
 #SBATCH --output={path}
 #SBATCH --constraint=intel
-#SBATCH --gres=gpu:1,vram:23G
-#SBATCH --partition=gpu_requeue
-#SBATCH -c 4
+#SBATCH --partition=gpu_quad
+#SBATCH --gres=gpu:1,vram:31G
+#SBATCH -c 8
 #SBATCH -n 1
-#SBATCH --mem=36GB
-#SBATCH --time=0-0:30:00
+#SBATCH --mem=48GB
+#SBATCH --time=0-7:30:00
 
 unset XDG_RUNTIME_DIR
 

face_rhythm/__init__.py

@@ -22,8 +22,8 @@
 
 
 ## Prepare cv2.imshow
-import pkg_resources
-installed_packages = {pkg.key for pkg in pkg_resources.working_set}
+import importlib.metadata
+installed_packages = {dist.metadata['Name'] for dist in importlib.metadata.distributions()}
 has_cv2_headless = 'opencv-contrib-python-headless' in installed_packages
 has_cv2_normal = 'opencv-contrib-python' in installed_packages
 if has_cv2_normal and not has_cv2_headless:
@@ -66,4 +66,4 @@ def prepare_cv2_imshow():
     exec('from . import ' + pkg)
 
 
-__version__ = '0.2.4'
+__version__ = '0.2.5'

face_rhythm/spectral_analysis.py

@@ -14,6 +14,20 @@ class VQT_Analyzer(FR_Module):
      displacement traces. The spectrograms are generated using
      the Variable Q-Transform (VQT) algorithm.
     RH 2022
+
+    Args:
+        params_VQT (dict):
+            A dictionary of parameters to pass to the VQT class (Variable
+            Q-Transform) in helpers.py.
+        normalization_factor (float):
+            A float between 0 and 1 to normalize the spectrograms. 0 means no
+            normalization. 1 means every time point has the same power.
+        spectrogram_exponent (float):
+            A float to raise the spectrogram to before normalizing.
+        one_over_f_exponent (float):
+            A float to raise the frequency axis to before doing 1/f correction.
+        verbose (int):
+            An integer to control the verbosity of the class.
     """
     def __init__(
         self,

face_rhythm/visualization.py

@@ -1,6 +1,7 @@
 from typing import Union
 from pathlib import Path
 import copy
+import gc
 
 import numpy as np
 import cv2
@@ -120,7 +121,7 @@ def __init__(
                 If list: Each element is a thickness for a different text.
                     Length of list must match the length of text.
         """
-        ## Stor arguments
+        ## Store arguments
         self.point_sizes = point_sizes if point_sizes is not None else None
         self.points_colors = points_colors if points_colors is not None else None
         self.alpha = alpha if alpha is not None else None
@@ -415,8 +416,11 @@ def visualize_image_with_points(
 
     def close(self):
         if self.video_writer is not None:
-            self.video_writer.release()
             cv2.destroyWindow(self.handle_cv2Imshow)
+            try:
+                self.video_writer.release()
+            except:
+                pass
 
     def __call__(self, *args, **kwds):
         """
@@ -426,11 +430,7 @@ def __call__(self, *args, **kwds):
         self.visualize_image_with_points(*args, **kwds)
     def __del__(self):
         self.close()
-    def __exit__(self):
-        self.close()
-    def __enter__(self):
-        return self
-
+
     def __repr__(self):
         return f'FrameVisualizer(handle_cv2Imshow={self.handle_cv2Imshow}, display={self.display}, video_writer={self.video_writer}, path_video={self.path_save}, frame_rate={self.frame_rate}, frame_height_width={self.frame_height_width})'
 
@@ -477,15 +477,26 @@ def play_video_with_points(
         ### Set buffered video reader to first frame
         bufferedVideoReader.set_iterator_frame_idx(int(idx_frames[0]))
         ### Iterate through frames
-        for idx_frame in tqdm(idx_frames):
-            frame = bufferedVideoReader[idx_frame][0]
-            frame = frame.numpy() if isinstance(frame, torch.Tensor) else frame
-            p = points_int[idx_frame] if points_int is not None else None
-            frameVisualizer.visualize_image_with_points(
-                image=frame,
-                points=[p],
-            )
-        frameVisualizer.close()
+        ### Use context manager to close frameVisualizer
+        class CM:
+            def __init__(self, frameVisualizer):
+                self.frameVisualizer = frameVisualizer
+            def __enter__(self):
+                return self.frameVisualizer
+            def __exit__(self, exc_type, exc_value, traceback):
+                self.frameVisualizer.close()
+                cv2.destroyWindow(self.frameVisualizer.handle_cv2Imshow)
+                gc.collect()
+        with CM(frameVisualizer) as f:
+            for idx_frame in tqdm(idx_frames):
+                frame = bufferedVideoReader[idx_frame][0]
+                frame = frame.numpy() if isinstance(frame, torch.Tensor) else frame
+                p = points_int[idx_frame] if points_int is not None else None
+                f.visualize_image_with_points(
+                    image=frame,
+                    points=[p],
+                )
+            f.close()
 
 
 # def display_toggle_image_stack(images, clim=None, **kwargs):

requirements.txt

@@ -6,7 +6,7 @@ notebook<7
 tensorly==0.8.1
 opencv_contrib_python==4.9.0.80
 matplotlib
-scikit_learn==1.4.1.post1
+scikit_learn==1.4.2
 scikit_image
 pyyaml
 tqdm