correct image ratio in pose detection and added sections_to_keep argument

Author: davidpagnon

Pose2Sim/Demo_Batch/Config.toml

@@ -190,6 +190,8 @@ interpolation = 'linear' #linear, slinear, quadratic, cubic, or none
                         # 'none' if you don't want to interpolate missing points
 interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 fill_large_gaps_with = 'last_value' # 'last_value', 'nan', or 'zeros' 
+sections_to_keep = 'all' # 'all', 'largest', 'first', 'last'
+                        # keep 'all' valid sections even when they are interspersed with untriangulated chunks, or the 'largest' valid section, or the 'first' one, or the 'last' one
 show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
 undistort_points = false # Better if distorted image (parallel lines curvy on the edge or at least one param > 10^-2), but unnecessary (and slightly slower) if distortions are low

Pose2Sim/Demo_Batch/Trial_1/Config.toml

@@ -189,6 +189,8 @@
                         # # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # fill_large_gaps_with = 'last_value' # 'last_value', 'nan', or 'zeros' 
+# sections_to_keep = 'all' # 'all', 'largest', 'first', 'last'
+                       #  # keep 'all' valid sections even when they are interspersed with untriangulated chunks, or the 'largest' valid section, or the 'first' one, or the 'last' one
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
 # undistort_points = false # Better if distorted image (parallel lines curvy on the edge or at least one param > 10^-2), but unnecessary (and slightly slower) if distortions are low

Pose2Sim/Demo_Batch/Trial_2/Config.toml

@@ -188,6 +188,8 @@ keypoints_to_consider = ['RWrist'] # 'all' if all points should be considered, f
 # interpolation = 'linear' #linear, slinear, quadratic, cubic, or none
                         # # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
+# sections_to_keep = 'all' # 'all', 'largest', 'first', 'last'
+                       #  # keep 'all' valid sections even when they are interspersed with untriangulated chunks, or the 'largest' valid section, or the 'first' one, or the 'last' one
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # fill_large_gaps_with = 'last_value' # 'last_value', 'nan', or 'zeros' 
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower

Pose2Sim/Demo_MultiPerson/Config.toml

@@ -189,6 +189,8 @@ interpolation = 'linear' #linear, slinear, quadratic, cubic, or none
                         # 'none' if you don't want to interpolate missing points
 interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 fill_large_gaps_with = 'last_value' # 'last_value', 'nan', or 'zeros' 
+sections_to_keep = 'all' # 'all', 'largest', 'first', 'last'
+                        # keep 'all' valid sections even when they are interspersed with untriangulated chunks, or the 'largest' valid section, or the 'first' one, or the 'last' one
 show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
 undistort_points = false # Better if distorted image (parallel lines curvy on the edge or at least one param > 10^-2), but unnecessary (and slightly slower) if distortions are low

Pose2Sim/Demo_SinglePerson/Config.toml

@@ -190,6 +190,8 @@ interpolation = 'linear' #linear, slinear, quadratic, cubic, or none
                         # 'none' if you don't want to interpolate missing points
 interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 fill_large_gaps_with = 'last_value' # 'last_value', 'nan', or 'zeros' 
+sections_to_keep = 'all' # 'all', 'largest', 'first', 'last'
+                        # keep 'all' valid sections even when they are interspersed with untriangulated chunks, or the 'largest' valid section, or the 'first' one, or the 'last' one
 show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
 undistort_points = false # Better if distorted image (parallel lines curvy on the edge or at least one param > 10^-2), but unnecessary (and slightly slower) if distortions are low

Pose2Sim/common.py

@@ -24,6 +24,7 @@
 import logging
 from anytree import PreOrderIter
 
+import tkinter as tk
 import matplotlib.pyplot as plt
 from matplotlib.backends.backend_qt5agg import FigureCanvasQTAgg as FigureCanvas
 from matplotlib.backends.backend_qt5agg import NavigationToolbar2QT as NavigationToolbar
@@ -1353,3 +1354,51 @@ def draw_keypts(img, X, Y, scores, cmap_str='RdYlGn'):
             if not (np.isnan(x[i]) or np.isnan(y[i]))]
 
     return img
+
+
+def get_screen_size():
+    '''
+    Get the screen dimensions
+
+    INPUTS:
+    - None
+
+    OUTPUTS:
+    - tuple of int: (screen_width, screen_height)
+    '''
+
+    root = tk.Tk()
+    screen_width = root.winfo_screenwidth()
+    screen_height = root.winfo_screenheight()
+    root.destroy()
+    
+    return screen_width, screen_height
+
+
+def calculate_display_size(W, H, screen_width, screen_height, margin=100):
+    '''
+    Calculate the optimal display size for the image
+    
+    INPUTS:
+        W, H: Original image dimensions
+        screen_width, screen_height: Screen dimensions
+        margin: Margin to leave around the window (pixels)
+    
+    OUTPUTS:
+        tuple: (display_width, display_height)
+    '''
+    
+    # If image fits within screen, use original size
+    if W <= screen_width - margin and H <= screen_height - margin:
+        return W, H
+    
+    # Calculate scaling factor to fit within screen while maintaining aspect ratio
+    width_ratio = (screen_width - margin) / W
+    height_ratio = (screen_height - margin) / H
+    scale_factor = min(width_ratio, height_ratio)
+    
+    # Calculate new dimensions
+    new_width = int(W * scale_factor)
+    new_height = int(H * scale_factor)
+    
+    return new_width, new_height

Pose2Sim/poseEstimation.py

@@ -46,7 +46,7 @@
 from rtmlib import PoseTracker, BodyWithFeet, Wholebody, Body, Hand, Custom, draw_skeleton
 from deep_sort_realtime.deepsort_tracker import DeepSort
 from Pose2Sim.common import natural_sort_key, sort_people_sports2d, sort_people_deepsort,\
-                        colors, thickness, draw_bounding_box, draw_keypts, draw_skel
+                        colors, thickness, draw_bounding_box, draw_keypts, draw_skel, get_screen_size, calculate_display_size
 from Pose2Sim.skeletons import *
 
 
@@ -233,7 +233,10 @@ def process_video(video_path, pose_tracker, pose_model, output_format, save_vide
         out = cv2.VideoWriter(output_video_path, fourcc, fps, (W, H)) # Create the output video file
 
     if display_detection:
-        cv2.namedWindow(f"Pose Estimation {os.path.basename(video_path)}", cv2.WINDOW_NORMAL + cv2.WINDOW_KEEPRATIO)
+        screen_width, screen_height = get_screen_size()
+        display_width, display_height = calculate_display_size(W, H, screen_width, screen_height, margin=50)
+        cv2.namedWindow(f"Pose Estimation {os.path.basename(video_path)}", cv2.WINDOW_NORMAL)
+        cv2.resizeWindow(f"Pose Estimation {os.path.basename(video_path)}", display_width, display_height)
 
     frame_idx = 0
     cap = cv2.VideoCapture(video_path)
@@ -353,7 +356,10 @@ def process_images(image_folder_path, vid_img_extension, pose_tracker, pose_mode
         out = cv2.VideoWriter(output_video_path, fourcc, fps, (W, H)) # Create the output video file
 
     if display_detection:
+        screen_width, screen_height = get_screen_size()
+        display_width, display_height = calculate_display_size(W, H, screen_width, screen_height, margin=50)
         cv2.namedWindow(f"Pose Estimation {os.path.basename(image_folder_path)}", cv2.WINDOW_NORMAL)
+        cv2.resizeWindow(f"Pose Estimation {os.path.basename(image_folder_path)}", display_width, display_height)
 
     f_range = [[0,len(image_files)] if frame_range in ('all', 'auto', []) else frame_range][0]
     for frame_idx, image_file in enumerate(tqdm(image_files, desc=f'\nProcessing {os.path.basename(img_output_dir)}')):

Pose2Sim/triangulation.py

@@ -669,6 +669,7 @@ def triangulate_all(config_dict):
     likelihood_threshold = config_dict.get('triangulation').get('likelihood_threshold_triangulation')
     interpolation_kind = config_dict.get('triangulation').get('interpolation')
     interp_gap_smaller_than = config_dict.get('triangulation').get('interp_if_gap_smaller_than')
+    sections_to_keep = config_dict.get('triangulation').get('sections_to_keep')
     fill_large_gaps_with = config_dict.get('triangulation').get('fill_large_gaps_with')
     show_interp_indices = config_dict.get('triangulation').get('show_interp_indices')
     undistort_points = config_dict.get('triangulation').get('undistort_points')
@@ -878,7 +879,7 @@ def triangulate_all(config_dict):
     # error_tot[0].to_csv(os.path.join(session_dir, 'error_tot.csv'), index=False, sep='\t')
 
     # Trim around good frames and remove persons with too few frames
-    f_range_trimmed = [indices_of_first_last_non_nan_chunks(err['mean'], min_chunk_size=interp_gap_smaller_than, chunk_choice_method='all') for err in error_tot]
+    f_range_trimmed = [indices_of_first_last_non_nan_chunks(err['mean'], min_chunk_size=interp_gap_smaller_than, chunk_choice_method=sections_to_keep) for err in error_tot]
     # f_range_trimmed = [f_range]*nb_persons_to_detect
     deleted_person_id = [n for n, f_range in enumerate(f_range_trimmed) if len(range(*f_range))<4]
     Q_tot = [Q_tot[n] for n in range(len(Q_tot)) if n not in deleted_person_id]

README.md

@@ -324,7 +324,7 @@ For example, try uncommenting `[project]` and set `frame_range = [10,99]`, or un
   Run it only when your cameras are moved or changed. If they are not, just copy a previous calibration.toml file into your new calibration folder.
 - `Pose2Sim.poseEstimation()`:
   - **Use your GPU**: This makes pose estimation significantly faster, without any impact on accuracy. See [Installation](#installation) section for more information.
-  - Set `det_frequency = 100` in Config.toml. Run the bounding box detector and the pose estimator on the first frame; for all subsequent frames, only run pose estimation: \
+  - Set `det_frequency = 100` in Config.toml. Run the bounding box detector and the pose estimator on the first frame; for all subsequent frames, only run pose estimation: 
   *150 s -> 30 s on my laptop with the Demo videos*
   - Use `mode = 'lightweight'`: Will use a lighter version of RTMPose, which is faster but less accurate\
   *30 s -> 20 s*